OneDroid Argus · live testing and audits of AI agents, workflows and fleets

Test the agents where they work: in the workflow, in the fleet, next to the people who sign off.

Code is not neutral. Neither is the agent running on it.

Every agent carries the biases of its training data, its prompt and whoever pays for it. A pass in a lab, on scenarios the vendor chose, with no human in the loop, says little about what the same agent does on your cases, strung into your workflow, next to your people. We test and audit agents in situ: what they did, what they claimed, what it cost, which human saw which decision, and what changes when one parameter changes.

The point is not fewer people. The work worth handing to agents is the work everyone hates: the copy-paste, the forwarded spreadsheets, the reports nobody reads, the waiting. The people keep the decisions and the contact with other people. OneDroid Argus tests whether that is still true once the agents move in: are the humans in the loop deciding, or rubber-stamping?

Today: Argus, automated end-to-end testing of AI-written code with checks the coding agent never sees, in use, its runner open source. In development: OneDroid for agentic workflows and fleets, in two editions: OneDroid Indie, one free, open-source binary on a developer's machine, and OneDroid Fleet, the record and the tests collected on your own estate.

The Argus runner runs inside your environment, and its source is open under MIT. No credit card.

Every checkable claim this site makes, and what backs it: onedroid.ai/evidence/onedroid.ai.

Testing agentic workflows and fleets · in development

Tests no lab can run for you.

An agent does not decide. It applies the rules a person gave it, and the biases it came with, and when a case falls outside them it still returns something that looks like an answer. A fleet test runs the same sealed scenarios against every agent you run and keeps the results on one record: regressions when one thing changes, edge cases the agent never saw, and whose interest the answer served.

The fleet: one sealed set, every agent.

Illustration: Four teams run the same sealed scenarios over six runs; after a model upgrade, the Finance team's runs start failing while the others keep passing.
The same sealed scenarios run against every team's agents on every run.

Regression: replay the run, change one thing.

Illustration: A recorded run of five steps and its replay with the model version changed; the replay's third step gives a different answer, and the replay falls outside the workflow's baseline.
A recorded run is replayed with one parameter changed: the model version, the prompt, a tool or a limit.

Red-teaming and edge cases: checks the agent never sees.

Illustration: A sealed set of edge-case scenarios held back from the agent; the agent receives only the task, and each scenario ends with one of four verdicts.
A tester writes the scenarios, they are sealed before the run, and the agent never sees what they expect: it gets the task, never the expected answer.

Bias analysis: whose interest did the answer serve?

Illustration: Left, an answer placed between the operator's, the vendor's and the user's interests; right, the same case run twice with only the postcode changed, giving two different answers.
Record whose interest each answer served, and run the same case with one detail changed: if the answer changes across repeated runs, that detail moved it.

Why we test in situ, and not in a lab.

Three things are documented. The fourth is not, and it is the one we test.

Feeds are tuned for the platform, and the tuning is what holds you. In a randomised experiment on Facebook and Instagram during the 2020 US election, moving users off the algorithmic feed "substantially decreased the time they spent on the platforms and their activity". Guess and colleagues, Science 381, 2023. The three-month effect on polarisation was null, so this shows the pull of the feed, not the harm of the bubble.

Platforms shape what you see to their own benefit. The European Commission fined Google EUR 2.42 billion in 2017 for giving "an illegal advantage to another Google product, its comparison shopping service". The US Federal Trade Commission's 2023 complaint alleges Amazon biased its search results "to preference Amazon's own products". European Commission, 27 June 2017; FTC, 26 September 2023, an allegation in a pending case.

"Free" agents have to earn, and their makers say so. Perplexity, announcing advertising: "subscriptions alone do not generate enough revenue", with ads "formatted as sponsored follow-up questions", and a promise that "the content of the answers you receive on Perplexity will not be influenced by advertisers". Perplexity, 12 November 2024. OpenAI announced ads in ChatGPT for free and Go users in January 2026, with a similar promise (SiliconANGLE, 16 January 2026).

Whether a deployed agent's answers favour its operator has not been measured. Every vendor promises the answer itself stays clean. We found no study that tested the promise on a running agent. That is a scenario, not an opinion: put the agent on the cases where the operator's interest, the vendor's and the user's diverge, and record whose interest the answers serve. The same goes for the biases nobody intended: they show at the edges, in the workflow, and only a test finds them.

A test the agent can see, it learns to pass. A test run in a lab, on the vendor's scenarios, with no human in the loop, measures the lab. So the checks are sealed before the run, held out from the agent, and run where the result is used, with the human's own act in the record. That is what Argus does for code today, and what OneDroid for agentic workflows and fleets is being built to do for everything else.

verdict log illustrative

    Holdout acceptance. Scenarios sealed before the run, against a pinned build.

    Under load. The same scenarios under load, against a message broker or an HTTP target; on an HTTP target a stepped ramp finds its limit.

    Continuously. Once it passes, it goes on a schedule, every 15 minutes at the shortest, and one page answers “Is it OK?”.

    Evidence anyone can replay. The withheld scenario set is hashed and written to a cryptographic ledger before the build starts. After the verdict the tester can reveal the set: every scenario is revealed — trigger, expectation, evidence hash, verdict — and the runner replays any of them against the same build. You never have to take our word for the verdict. How the seal and the custody work.

    Argus · the code testing product, available today

    50,000 lines overnight. How do you know what it did?

    Nobody reads all of it. You can hold it to what it was supposed to do, with checks it never saw. Two things make that answer worth trusting.

    The builder can't game checks it never sees. A model that can see its checker learns to satisfy the checker. OneDroid Argus seals the checks before the work starts. The builder, a person or an AI agent, holds a credential that can ask for a run and read the verdict, and cannot read a check or an expected value.

    Every verdict comes with its boundary. The reveal lists every check that was run. The certificate says what it proves and what it does not, for instance whether the running build was measured. Our case studies also list what was not tested. A pass without its boundary is only a claim.

    Shown on real code: a decades-old Fortran library held to its own user guide, 100 sealed checks, each quoting the sentence it tests. 91 held, 9 failed, the checks that depend on the changed constants were shown able to fail, and the page lists what the result does not prove.

    It solves the problem, and it is affordable. In one measured experiment the coding agent built a feature for $1.01 to $1.97 a run, and the independent tester was one session of about $5.30 for the whole feature. Independent testing now costs the same order as the AI coding.

    You already had the incident.

    80% have traced a production incident directly back to AI-generated code, and 92% aren't confident their current safeguards would catch an AI-driven failure before it reaches users. Sauce Labs, “The Enterprise AI Code Verification Crisis”, n = 400 US executives and engineering leaders, 22 July 2026 — a vendor-commissioned survey. (AS-9 §2, fetched 2026-09-10.)

    3.1% of developers highly trust AI output; 19.6% highly distrust it; 66% hit answers that are “almost right, but not quite”. Stack Overflow Developer Survey 2025, n = 33,244. (AS-9 §2, fetched 2026-09-10.)

    Syntax pass rates for AI-generated code climbed from about 50% to 95%. Security pass rates stayed flat, between 45% and 55%. Veracode, Spring 2026 GenAI Code Security update, 150+ models across 80 coding tasks, 24 March 2026. (AS-9 §2, fetched 2026-09-10.) Capability is improving. Correctness under adversarial check is not.

    And the tools that read the diff have stepped back from the verdict on purpose: Claude Code Review's check “always completes with a neutral conclusion so it never blocks merging”; Copilot's reviews “do not count toward required approvals”. (Vendor documentation, AS-9 §1 and §5, fetched 2026-09-10.) OneDroid Argus returns a verdict: passed or failed, or not_applicable when no check could run. A check that could not run never counts as a pass; a load run can be degraded; a comparison run reports differences, not a pass.

    How it works

    Three steps, and the middle one is the point.

    01

    Your agent writes the scenarios

    You state the intent — the ticket, the acceptance criteria, the interface diff. Your own agent, with our skill, turns that into a scenario set: the happy path, the repeat, the wrong actor, the wrong state, every external call's failure, the race at the shared resource, and the load variant. The control plane seals the set and anchors its hash before the run.

    02

    The builder never sees the answer key

    The coding agent never holds a token that can read the expectations, and the scenarios are never in its working directory. It builds against reality, not an answer key. It can run the scenarios as often as it likes and see only what happened — never what was expected.

    03

    The verdict, with evidence

    One final run against a pinned build, then the reveal: every scenario, its trigger, its expectation, the evidence hash, and the verdict — as a report a reviewer can hand to an auditor, and a certificate anyone can check with the open CLI and no OneDroid account. The scenarios become regression guards; the next feature gets a fresh withheld set.

    Three modes, one evidence contract

    One product, three questions.

    Holdout acceptance — “does it do what was asked?”

    Withheld scenarios, run once against a pinned artefact: a verdict plus per-scenario replayable evidence.

    Under load — “how much can it take?”

    A scenario declares its load and the latency and error rate it must stay under. For message brokers, a ramp of sessions in steps finds the limit and records what was sent, confirmed and delivered at each step. An HTTP target can be ramped step by step too. No ramp drives a web page in a browser.

    Continuously — “is it still true tomorrow?”

    The reviewed scenario set runs on a monitor schedule, every 15 minutes at the shortest. One page answers “Is it OK?” from the scheduled runs only, and lists what needs a person.

    All three in detail, and what OneDroid Argus is not.

    The run layer

    It brings its own run layer.

    OneDroid Synapse is the governed MCP gateway. Your agents connect to one URL instead of holding credentials for each service; every tool call is authenticated, policy-checked, and logged while the hub's activity logging is on (fail-open: a logging failure never blocks a call), to an append-only audit log you own. It ships with OneDroid Argus and runs standalone. How Synapse works

    OneDroid Engram is versioned knowledge and memory in a Postgres you control with your own database, or one OneDroid hosts — hybrid vector, keyword and graph search, git-style history, diff and restore. How Engram works

    The method behind the gate is written down and open: the Dark Factory kits, Apache-2.0.

    For regulated buyers

    When someone has to answer for it.

    OneDroid Argus, Synapse and Engram on your own infrastructure, held together by one thing: the evidence record. Synapse's audit entry, Engram's versioned fact and OneDroid Argus's verdict are the same object — a promise plus the evidence for it. Where the law names the pieces they line up: EU AI Act Art. 12 (automatic recording of events over the lifetime of the system) and Art. 14 (human oversight) for high-risk AI products; DORA Art. 25 and Art. 30; HIPAA 164.312(b); FDA 21 CFR 11.10(e). Using a coding agent does not by itself make your system high-risk, and we will not tell you otherwise.

    Worth knowing: the largest vendor's hosted agents say of themselves that they are “not currently eligible for Zero Data Retention (ZDR) or HIPAA Business Associate Agreement (BAA) coverage.” (Vendor documentation, AS-10 via strategy §6b, fetched 2026-09-10.) If that is your world, start here.

    Open source

    Open where it has to be.

    github.com/OneDro1d open source

    The Dark Factory kits are open source under Apache-2.0, at the address above. The OneDroid Argus runner executes inside your environment and touches your data plane, so its source is open too, under MIT: github.com/OneDro1d/argus-runner. Your tester writes the scenarios with our open skill; the control plane that seals the set and adjudicates the verdict is hosted and closed. The holdout is provably prior, and the evidence is replayable with the runner in your own environment. A certificate proves what the runner reported and when, not that the runner was unmodified.

    FAQ

    Questions people actually ask.

    What is OneDroid Argus?

    Automated end-to-end testing for AI-written code. It runs acceptance scenarios your coding agent never saw against the system it built, under load as well as once, and returns a verdict with evidence you can replay.

    Is OneDroid Argus end-to-end testing?

    Yes, and it does more than a usual end-to-end suite. Like any end-to-end test it exercises the running system, with nothing mocked: HTTP, the message broker, the database, web pages in a real browser. Four things differ. Checks are plain scenarios, not test scripts: a separate tester, your QA engineer or an agent but never the builder, writes them from the ticket and the acceptance criteria, before the build. There are no test scripts to write and keep alive, except a spec file for a browser check: the scenarios are written once per feature, before the build, and stay as regression checks for the next change. Your coding agent can't pass by guessing, because its credential cannot read a check. And the same checks run under load (a message broker or an HTTP target) and on a schedule, every 15 minutes at the shortest, with no second tool.

    How is this different from an AI code reviewer?

    A reviewer reads the diff and gives an opinion. OneDroid Argus does not judge the diff; it exercises the running system. A reviewer says “this looks wrong”; OneDroid Argus says “when I cancelled a shipped order, you refunded it”, and shows you the saga.

    What does independent testing cost?

    The same order as the coding. In one measured experiment, Claude Code built a feature for $1.97 a run testing its own work; with OneDroid Argus as the independent tester, the coding agent cost $1.01 a run and the tester one session of about $5.30, which wrote both the build checks and the 13 hidden judge checks for the feature, and both setups passed all 13 hidden checks. The tester is paid once per feature, and its checks stay as regression checks. One ticket and three runs per setup, so treat it as one data point. The case study.

    Do I have to trust your closed control plane?

    No. The scenario set is hashed and anchored before the build, revealed in full, with an on-chain anchor, when the tester chooses to reveal after the verdict, and any revealed scenario can be replayed by the runner. The anchor proves the scenarios were prior and unchanged; keeping them unseen is custody — a builder credential that cannot read a test, and an environment that does not hold one.

    Where does it run?

    Next to your system, inside your environment, and on the owner's terms. A monitor can watch a live system; for an app that moves money, one switch makes OneDroid Argus refuse every write the owner has not listed, with limits. The chain check reads your database in a read-only transaction. A database-state check runs the VERIFY query you write; it is screened for write keywords, and the configuration guide asks for a read-only database role. OneDroid Argus writes to it only if a scenario declares a cleanup block, with the credential you give it. A load ramp runs only against a target its operator listed, with production-looking addresses refused.

    Does OneDroid Argus score my build?

    No. Every scenario ends as passed, failed, errored or not_applicable (it could not run; a load run has one more outcome, degraded), and neither an error nor not_applicable is ever a pass. The verdict is a decision about one pinned build against the sealed scenarios, with the evidence attached. There is no number to optimise for and no “safe” label.

    Find out what your agent actually built.

    The OneDroid Argus runner is free and open source under MIT; the control plane: sign up to evaluate, your own environment by request. Email michal@onedroid.ai — you'll reach Michal Bacia, the team that builds the product, not a sales queue. We reply within one business day.

    Every checkable claim this site makes, and what backs it

    672 claims as of 2026-11-07: 304 evidenced, 12 contradicted (11 of them low confidence), 329 with no evidence published, 27 with evidence available on request. The same verdicts, rules and limits as on the full page, which also carries the change log and how to correct a line. Nothing here says a claim is true or false; every reference points into a source whose sha256 as fetched is listed at the end.

    Contradicted (1)

    IdClaim, as madeMade atBoth sides
    C-214Every study comes in two versions with the same facts: the test page for engineers, and a blog post for leaders and product owners.onedroid.ai/llms-full.txt (sources/llms-full.txt L3126-3127 (PAGE /case-studies; under "What happened when"))/case-studies lists five studies and says every one has a blog version, but the blog index lists four case-study posts and none for 'Three ways to test code you did not write', whose page has no business-view link (the other studies have one at L24). The new study replaced code-vs-user-guide in round 7.
    sources/llms-full.txt#L3126-3127
    sources/llms-full.txt#L3131-3171
    https://onedroid.ai/blog/llms.txt#200

    Contradicted, low confidence (11)

    The disprove pass itself said a charitable reading reconciles each of these: a dated historical entry, a paraphrase, a feature documented for the hosted service but not yet in the public code snapshot. They are listed because the two surfaces do disagree as written.

    IdClaim, as madeMade atBoth sides
    C-012**Faster and cheaper than end-to-end testing by hand.** No test scripts to write and keep alive (except a spec file for a browser check), and nobody clicking through the app before each release.docs.onedroid.ai/argus.md (sources/argus-docs.md L23-24 (under "Is it end-to-end testing?"))Low confidence: docs and /argus say no test scripts 'except a spec file for a browser check', while the round-seven home FAQ and llms.txt say flatly 'There are no test scripts to write and keep alive', dropping the exception; a charitable reading treats the FAQ as a summary. The 'faster and cheaper than by hand' half has no published measurement.
    sources/argus-docs.md#L23-24
    sources/argus.md#L47
    sources/llms-full.txt#L287
    C-050It comes with a certificate: proof without the tests, which anyone can check with the open CLI, with no OneDroid account and no contact with OneDroid: it reads the public chain through an RPC endpoint you choose.onedroid.ai/llms-full.txt (sources/llms-full.txt L2848-2851 (PAGE /argus; under "The reveal"))
    onedroid.ai/argus.md (sources/argus.md L112-115 (under "The reveal"))
    Low confidence: /argus and the README say verify reads the public chain through an RPC endpoint you choose (--rpc), while the partners page says a certificate is one 'anyone can check offline with the CLI and no account'; 'offline' may mean 'without OneDroid'.
    sources/argus.md#L112-115
    sources/argus-runner-README.md#L107-112
    sources/partners-financial-services.html#L311
    C-069OneDroid Argus is in early use, and compare is its newest part.onedroid.ai/llms-full.txt (sources/llms-full.txt L3176 (PAGE /case-studies; under "Six differences between two releases of a ticketing system"))Low confidence: /case-studies and /argus call compare the newest part, but docs date compare to v0.3.57, and HTTP load ramps (v0.3.66), fault steps (0.3.69), db/logs steps (0.3.70) and the bus step (0.3.74) came later; 'part' may mean a mode, not a feature.
    sources/llms-full.txt#L3176
    sources/argus.md#L169-170
    sources/argus-docs.md#L225
    C-084prints the literal fix for each problem. It writes nothing except a renewed session token, and never prints a credential.docs.onedroid.ai/argus-quickstart (sources/argus-quickstart.html L234-235 (under "4. Prove read access before wiring anything else"))
    docs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L233-234 (under "4. Prove read access before wiring anything else"))
    Low confidence: the quickstart (and doctor.go, which saves a rotated refresh token to the session file) say doctor writes a renewed session token, while the runner README twice calls argus doctor 'read-only'; 'read-only' may mean towards the system and config.
    sources/argus-quickstart.md#L233-234
    sources/argus-runner-README.md#L46-47
    sources/argus-runner-README.md#L96-97
    C-176Checks read your database in read-only transactions; OneDroid Argus writes to it only if a scenario declares a cleanup block, and then only with the credential you give it.onedroid.ai/llms-full.txt (sources/llms-full.txt L2970 (PAGE /argus; under "Not a free hand on a live system"))
    onedroid.ai/argus.md (sources/argus.md L234 (under "Not a free hand on a live system"))
    Low confidence: the chain db step runs in BEGIN READ ONLY (pgread.go L108), but a Database State check's VERIFY query is sent to JMeter's JDBC sampler with autocommit true and no read-only setting, so that kind of check does not read in a read-only transaction. A charitable reading reconciles it: the query is linted for write keywords and the config guide asks for a READ-ONLY role.
    sources/argus.md#L234
    repo/templates/database-state.jmx#L45
    repo/templates/database-state.jmx#L227-235
    C-248No test scripts to write and keep alive (except a spec file for a browser check), and nobody clicking through the app before each release.docs.onedroid.ai/argus (sources/argus-docs.html L74-75 (under "Is it end-to-end testing?"))
    docs.onedroid.ai/argus.md (sources/argus-docs.md L23-24 (under "Is it end-to-end testing?"))
    Low confidence. Docs, /argus and /qa-teams give a browser-check spec file as the one exception to 'no test scripts', but the home FAQ and llms.txt still say flatly 'There are no test scripts to write and keep alive', while also listing web pages in a real browser; charitably the FAQ is a shorter summary.
    sources/argus-docs.md#L23-24
    sources/llms-full.txt#L2783
    sources/llms-full.txt#L287
    C-262A run needs an account on the hosted control plane.onedroid.ai/llms-full.txt (sources/llms-full.txt L2941 (PAGE /argus; under "MCP server"))
    onedroid.ai/argus.md (sources/argus.md L205 (under "MCP server"))
    Low confidence. /argus says 'A run needs an account on the hosted control plane', while the runner README says whoever holds a reveal can 're-run any scenario in it, without a OneDroid account' with `argus replay`; charitably, a run means a control-plane run and a replay is something else.
    sources/llms-full.txt#L2941
    sources/argus.md#L205
    sources/argus-runner-README.md#L119-125
    C-515No test scripts to write and keep alive, except a spec file for a browser checkonedroid.ai/llms-full.txt (sources/llms-full.txt L2742 (PAGE /argus))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L2783 (PAGE /argus; under "Not a test suite to maintain"))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L5181 (PAGE /qa-teams; under "Plain scenarios, not test scripts"))
    /argus and /qa-teams say 'No test scripts to write and keep alive, except a spec file for a browser check', while the home FAQ (llms-full L287) and llms.txt L42 still say flatly 'There are no test scripts to write and keep alive'. Low confidence: the FAQ may be a shorter summary.
    sources/llms-full.txt#L2783
    sources/argus.md#L47
    sources/llms-full.txt#L287
    C-534It lives in a private GitHub repository you own.onedroid.ai/llms-full.txt (sources/llms-full.txt L5501 (PAGE /kit; under "The method: the Dark Factory kits"))/kit says of the open-source method 'It lives in a private GitHub repository you own', yet the same page puts the runbook 'in the public repository' and clones github.com/OneDro1d/dark-factory, which /quickstart calls '(public, Apache-2.0)'. Low confidence: 'It' may mean the human's own copy, which START-HERE puts in a private repo.
    sources/llms-full.txt#L5500-5501
    sources/llms-full.txt#L5591-5593
    sources/llms-full.txt#L4545-4546
    C-578It comes with a certificate anyone can check offline with the CLI and no account.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L311 (under "02What OneDroid Argus does"))Low confidence. The partners page says the certificate can be checked 'offline', but /argus and the runner README say verify 'reads the public chain through an RPC endpoint you choose' (--rpc). A charitable reading takes 'offline' to mean 'without OneDroid', which is what the code comment (OFFLINE and INDEPENDENT, never calls the control plane) means.
    sources/partners-financial-services.html#L311
    sources/argus.md#L113-115
    sources/argus-runner-README.md#L107-110
    C-718Every run had a written prediction beforehand and matched it. The runs weren't sealed.onedroid.ai/llms-full.txt (sources/llms-full.txt L3896 (PAGE /case-study-three-ways-to-test-code; under "What this does not show"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L81 (under "What this does not show"))
    'The runs weren't sealed' (L81; llms.txt 'not sealed'), yet the same page's section 1 bundle holds 'the certificate, the final and rehearsal reports, the sealed checks' (L83). Low confidence: 'runs' may mean the predictions or the result rather than the checks.
    sources/case-study-three-ways-to-test-code.md#L81
    sources/case-study-three-ways-to-test-code.md#L83
    sources/llms-full.txt#L3896

    Evidenced (304)

    IdClaim, as madeMade atArtefact
    C-026Seven checks sealed before the run, against the FINOS CALM trades API and MCP demo: 5 passed, 2 failed, 0 errored.onedroid.ai/llms-full.txt (sources/llms-full.txt L3379 (PAGE /case-study-finos-calm))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L4 (under "Does the guardrail hold on every path? Seven sealed checks against a FINOS CALM demo — OneDroid Argus case study"))
    https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#200 block_number 47817413, timestamp 2026-10-07T19:58:34Z, input payload scenario_count 7
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#200 block_number 47817439, input payload verdict failed, tallies passed 5 failed 2 errored 0
    sources/case-study-finos-calm.md#L131-133
    The public Base Sepolia transactions the page names record a 7-scenario set sealed before the run and a verdict of passed 5, failed 2, errored 0 after it; PARTIAL: that the target was the FINOS CALM trades API and MCP demo is supported only by the anchored artifact_digest sha256:feb820a3... matching the page's trades-mcp-server digest prefix (L57).
    C-030The sealed set, before the run (block 47817413) | `0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc` [...] The verdict, after the run (block 47817439) | `0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a`onedroid.ai/llms-full.txt (sources/llms-full.txt L3504-3506 (PAGE /case-study-finos-calm; under "How anyone can check it later"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L131-133 (under "How anyone can check it later"))
    https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#200 block_number 47817413, status ok, payload set_hash/holdout_root/criteria_hash
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#200 block_number 47817439, status ok, payload run_id 20261007T195848022 verdict failed
    sources/case-study-finos-calm.md#L131-133
    Both quoted transactions exist on Base Sepolia at exactly the quoted blocks, the first carrying the sealed set's hashes and the second the run's verdict, chained by previousCorrelationId.
    C-041Like any end-to-end test, OneDroid Argus exercises the running system, with nothing mocked: HTTP, the message broker, the database, web pages in a real browser.onedroid.ai/llms-full.txt (sources/llms-full.txt L2773-2774 (PAGE /argus; under "Yes, and it does more than a usual end-to-end suite."))
    onedroid.ai/argus.md (sources/argus.md L37-38 (under "Yes, and it does more than a usual end-to-end suite."))
    repo/README.md#L8
    repo/internal/argus/ui_scenario.go#L37-38
    repo/internal/scenario/validate.go#L474
    The open runner has HTTP, read-only database, AMQP broker and Playwright web-UI scenario types; PARTIAL: 'nothing mocked' is a property of use, not shown by an artefact.
    C-044Its credential can ask for a run and read the verdict, and cannot read a check or an expected value.onedroid.ai/llms-full.txt (sources/llms-full.txt L2787 (PAGE /argus; under "Your coding agent can't game it."))
    onedroid.ai/argus.md (sources/argus.md L51 (under "Your coding agent can't game it."))
    repo/internal/auth/auth.go#L8-9
    repo/internal/mcpserver/tools.go#L34-77
    repo/internal/mcpserver/tools.go#L60
    The product hat is limited to runner_* tools (run, get_report, sagas, logs) and get_report strips expected values for it; PARTIAL: enforcement on the hosted control plane's credential is not visible in the open code.
    C-045During the build it reports only what happened — failing ids and observed reality, so your agent can fix things — never what was expected.onedroid.ai/llms-full.txt (sources/llms-full.txt L2838-2839 (PAGE /argus; under "Run unseen, then adjudicated."))
    onedroid.ai/argus.md (sources/argus.md L102-103 (under "Run unseen, then adjudicated."))
    repo/internal/mcpserver/tools.go#L60
    repo/internal/report/report.go#L248
    repo/internal/report/report.go#L300
    The report redacts the expected value for the product hat (redactExpected) while keeping observed traffic, matching 'observed reality, never what was expected'.
    C-046The scenario set is hashed and written to a cryptographic ledger before the build starts and revealed in full, with an on-chain anchor, when the tester chooses to reveal after the verdictonedroid.ai/llms.txt (sources/llms.txt L32-34 (under "OneDroid"))https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#200 block 47817413 payload set_hash, holdout_root, scenario_count 7
    repo/internal/reveal/document.go#L90-106
    repo/README.md#L119-121
    A set hash anchored on a public chain before the verdict anchor, plus a reveal document format with sealed files and Merkle proofs; PARTIAL: no revealed set is publicly published, and 'before the build starts' (vs before the run) is not shown.
    C-048The final run is made against a pinned artefact, and the control plane adjudicates it.onedroid.ai/llms-full.txt (sources/llms-full.txt L2839 (PAGE /argus; under "Run unseen, then adjudicated."))
    onedroid.ai/argus.md (sources/argus.md L103 (under "Run unseen, then adjudicated."))
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#200 payload artifact_digest sha256:feb820a3...
    repo/internal/certificate/certificate.go#L25-29
    The anchored verdict binds a pinned artifact digest and the v3+ certificate requires an artifact measurement; PARTIAL: that the control plane adjudicates is not visible (README L25-27 says it signs and anchors what the runner reports).
    C-049The verdict binds to a specific build, so it cannot be re-pointed at a later one.onedroid.ai/llms-full.txt (sources/llms-full.txt L2848 (PAGE /argus; under "The reveal"))
    onedroid.ai/argus.md (sources/argus.md L112 (under "The reveal"))
    onedroid.ai/partners-financial-services (sources/partners-financial-services.html L310 (under "02What OneDroid Argus does"))
    repo/internal/certificate/certificate.go#L25-29
    repo/README.md#L131-133
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#200 artifact_digest field
    The verdict anchor carries the artifact digest, verify checks it, and replay refuses any other digest (README L132-133).
    C-051Because the runner is open source, a certificate proves what the runner reported and when, not that the runner was unmodified.onedroid.ai/llms-full.txt (sources/llms-full.txt L2851-2852 (PAGE /argus; under "The reveal"))
    onedroid.ai/argus.md (sources/argus.md L115-116 (under "The reveal"))
    repo/LICENSE#L1
    repo/README.md#L25-32
    The runner is MIT open source (LICENSE) and the repo README states the same limit, that verification proves what was reported and anchored, not an unmodified runner.
    C-052Decoy checks, whose purpose is to catch a runner that reports results it did not earn, **are not implemented yet**raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L29-30 (under "What a OneDroid Argus certificate proves today, and what it does not"))repo/README.md#L29-30
    repo/internal/router/runner_holdout_test.go#decoy (test only)
    repo/internal/argus/saga_presence_judge_test.go#decoy (test only)
    A grep of the repo for 'decoy' finds only the README and two test files, with no implementation, consistent with 'not implemented yet'.
    C-053an anchor proves the statement existed no later than its block; it does not prove the statement is true. A Bitcoin (OpenTimestamps) receipt is "pending" until the calendar has a block attestation.raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L34-36 (under "What a OneDroid Argus certificate proves today, and what it does not"))repo/README.md#L34-37
    repo/internal/certificate/verify.go#L38
    repo/internal/certificate/verify.go#L607
    The verifier has a 'pending' status that it gives to an OpenTimestamps receipt without a block attestation; the 'existed no later than its block' semantics is stated in the README.
    C-054Exit code 0 means no check mismatched and at least one of the run's verdict anchors read back and matched.raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L112-113 (under "Verify a certificate"))repo/internal/certificate/verify.go#L681-695
    repo/cmd/argus/certificate_cmd.go#L210-213
    Overall() returns false on any mismatch and needs a verified verdict anchor, and the command exits OK only when it is true.
    C-055When the tester chooses to reveal the set, every scenario is published with its trigger, expectation, evidence hash and verdict, and any scenario can be replayed by the runner against the same artefactonedroid.ai/llms-full.txt (sources/llms-full.txt L2852-2854 (PAGE /argus; under "The reveal"))
    onedroid.ai/argus.md (sources/argus.md L116-118 (under "The reveal"))
    repo/internal/reveal/document.go#L90-106
    repo/README.md#L119-133
    The reveal document format carries sealed files, proofs, verdicts and the evidence bundle hash, and `argus replay` re-runs a revealed scenario against the same digest; PARTIAL: no actual reveal is publicly published.
    C-056Every scenario ends as passed, failed, errored or not_applicable (it could not run), and neither an error nor not_applicable is ever a pass.onedroid.ai/llms-full.txt (sources/llms-full.txt L2858 (PAGE /argus; under "The reveal"))
    onedroid.ai/argus.md (sources/argus.md L122 (under "The reveal"))
    onedroid.ai/partners-financial-services (sources/partners-financial-services.html L322 (under "02What OneDroid Argus does"))
    repo/internal/certificate/verify.go#L152-193
    repo/internal/runner/mapper.go#L110-116
    The verifier tallies passed/failed/errored/not_applicable separately and refuses a 'passed' verdict with errors or with only not_applicable ('never a pass', L189); note that mapper.go L114 rolls a run with some not_applicable and the rest passed up to 'passed'.
    C-057It is not a score: there is no number to optimise for and no “safe” label.onedroid.ai/llms-full.txt (sources/llms-full.txt L2860-2861 (PAGE /argus; under "The reveal"))
    onedroid.ai/argus.md (sources/argus.md L124-125 (under "The reveal"))
    repo/internal/certificate/certificate.go#L18-40
    repo/docs/CERTIFICATE.md
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#200 payload verdict + tallies only
    Neither the certificate format nor docs/CERTIFICATE.md has any 'score' or 'safe' field (grep: no match), and the anchored verdict payload carries only verdict and tallies.
    C-058OneDroid Argus returns a verdict: passed or failed, or not_applicable when no check could run.onedroid.ai/llms-full.txt (sources/llms-full.txt L165 (PAGE /; under "You already had the incident."))repo/internal/certificate/verify.go#L377-378
    repo/internal/certificate/verify.go#L191-193
    The verdict vocabulary has passed, failed and not_applicable, the last only when every scenario was not applicable; PARTIAL: the format also has 'errored' and 'degraded' tallies the sentence does not mention.
    C-059A scenario can declare a load: a number of users held for a duration, with the p95 latency and the error rate it must stay under. For message brokers there is more: a ramp of sessions in steps, up to 2,000 sessions a step by default, that finds the limit.onedroid.ai/llms-full.txt (sources/llms-full.txt L2867-2869 (PAGE /argus; under "Under load."))
    onedroid.ai/argus.md (sources/argus.md L131-133 (under "Under load."))
    repo/internal/scenario/amqpload.go#L161-162
    repo/internal/scenario/amqpload.go#L32
    repo/internal/amqpload/amqpload.go#L44
    Load scenarios declare steps, target p95 and max error rate, and AMQP Steps are bounded to [1, 2000] sessions per step.
    C-060Each step records the rates offered, sent, confirmed and delivered, the time from publish to confirm and to delivery, the errors by class, and any time the broker blocked.onedroid.ai/llms-full.txt (sources/llms-full.txt L2869-2871 (PAGE /argus; under "Under load."))
    onedroid.ai/argus.md (sources/argus.md L133-135 (under "Under load."))
    repo/internal/amqpload/amqpload.go#L331-341
    repo/internal/amqpload/amqpload.go#L265-293
    repo/internal/amqpload/amqpload.go#L454
    Each AMQP step records offered/sent/confirmed/delivered rates, errors, blocked periods with reason and kind, and publish-to-deliver latency.
    C-061An HTTP target can be ramped too: users are raised step by step against one named target, and the ramp stops at the first step that is not comfortable, so a system that has broken is not pushed harder.onedroid.ai/llms-full.txt (sources/llms-full.txt L2873-2875 (PAGE /argus; under "Under load."))
    onedroid.ai/argus.md (sources/argus.md L137-139 (under "Under load."))
    repo/internal/scenario/httpload.go#L16
    repo/internal/argus/httpload_run.go#L28
    repo/internal/httpload/httpload.go#L201
    The HTTP ramp code stops at the first step that is not comfortable ('so a target that has broken is not pushed harder').
    C-062The limits, stated plainly: stepped ramps exist for AMQP brokers and for one HTTP target at a time, and none drives a web page in a browser.onedroid.ai/llms-full.txt (sources/llms-full.txt L2878-2879 (PAGE /argus; under "Under load."))
    onedroid.ai/argus.md (sources/argus.md L142-143 (under "Under load."))
    repo/internal/scenario/httpload.go#L16
    repo/internal/scenario/amqpload.go#L161-162
    repo/onboarding/HOW-TO-ARGUS-CONFIG.md#L139-140
    Stepped ramps exist in code for AMQP (one named broker) and HTTP (one named target); PARTIAL: that no ramp drives a browser page is an absence, not shown by an artefact.
    C-064A load ramp runs only against a target its operator listed, and an address that looks like production is refused.onedroid.ai/llms-full.txt (sources/llms-full.txt L2970 (PAGE /argus; under "Not a free hand on a live system"))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L4637 (PAGE /security; under "Where the verifier runs, and where it never does."))
    onedroid.ai/argus.md (sources/argus.md L234 (under "Not a free hand on a live system"))
    repo/internal/config/config.go#L1078-1081
    repo/internal/config/config.go#L1000-1011
    repo/onboarding/HOW-TO-ARGUS-CONFIG.md#L139-140
    Load runs only against load_allowed_targets entries, and a shared or production-looking broker is refused at config load (validateLoadAllowed, looksSharedOrProd).
    C-068The result is a table: rows are checks, columns are systems, and each cell reads same, differs, approved, noise, not measured or could not run. What may differ is declared before the run, with a tolerance for numbers. A difference a person accepts is pinned to the two outputs they were shownonedroid.ai/llms-full.txt (sources/llms-full.txt L2899-2902 (PAGE /argus; under "One sealed set, two systems."))
    onedroid.ai/argus.md (sources/argus.md L163-166 (under "One sealed set, two systems."))
    repo/internal/compare/judge.go#L8-11
    repo/internal/compare/result.go#L9
    repo/internal/compare/result.go#L225-229
    The compare code defines cell states same, differs, approved, noise, not_measured and could_not_run; PARTIAL: tolerances and pinning an approval to the two outputs shown were not traced line by line.
    C-070Between steps a check can read one row of the system's PostgreSQL database with a read-only query, or ask its log store (Loki) whether a line was written. A check can also make the system fail on purpose while it runs: stop, start or scale a deployment, or delay or cut one connection.onedroid.ai/llms-full.txt (sources/llms-full.txt L2841-2844 (PAGE /argus; under "Run unseen, then adjudicated."))
    onedroid.ai/argus.md (sources/argus.md L105-108 (under "Run unseen, then adjudicated."))
    repo/internal/scenario/validate.go#L474
    repo/internal/scenario/chain.go#L162-168
    repo/docs/fault-steps.md#L3
    The chain grammar has a read-only DB query step, a Loki logs step, and fault steps (stop, scale, delay or cut a link).
    C-071Nothing can be made to fail unless the system's operator has listed it; OneDroid Argus undoes every fault it injected, and the report and the certificate say what was done and whether every undo succeeded.onedroid.ai/llms-full.txt (sources/llms-full.txt L2844 (PAGE /argus; under "Run unseen, then adjudicated."))
    onedroid.ai/argus.md (sources/argus.md L108 (under "Run unseen, then adjudicated."))
    repo/docs/fault-steps.md#L8
    repo/docs/fault-steps.md#L126-131
    repo/internal/certificate/certificate.go#L31-35
    No fault runs without fault_allowed_targets, every active fault is undone with restore_outcome recorded, and the v4 certificate lists the applied faults.
    C-073For an app that moves money, one switch makes OneDroid Argus refuse every write the owner has not listed, with limits.onedroid.ai/llms-full.txt (sources/llms-full.txt L2968-2970 (PAGE /argus; under "Not a free hand on a live system"))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L4636-4637 (PAGE /security; under "Where the verifier runs, and where it never does."))
    onedroid.ai/argus.md (sources/argus.md L232-234 (under "Not a free hand on a live system"))
    repo/internal/config/config.go#L331-338
    repo/internal/config/config.go#L1101-1105
    A money_handling switch with a money_writes allow-list is in the config and validated at load.
    C-074The runner executes inside your own environment next to the system under test.onedroid.ai/llms-full.txt (sources/llms-full.txt L4639-4640 (PAGE /security; under "Where the verifier runs, and where it never does."))
    onedroid.ai/security.md (sources/security.md L30-31 (under "Where the verifier runs, and where it never does."))
    repo/README.md#L8
    repo/README.md#L43-44
    repo/README.md#L80-83
    The runner is deployed by render-k8s into the user's own cluster next to the system under test and connects out only.
    C-075What leaves your boundary is the evidence the control plane needs to adjudicate a verdict: statuses, timings and, unless the tester asks for them to be cut, short excerpts of what the application returned; not your data stores.onedroid.ai/llms-full.txt (sources/llms-full.txt L4642-4643 (PAGE /security; under "Where the verifier runs, and where it never does."))
    onedroid.ai/security.md (sources/security.md L33-34 (under "Where the verifier runs, and where it never does."))
    repo/internal/runner/mapper.go#L140
    repo/internal/chain/body_excerpt.go#L11-39
    The results push carries summaries with no expected or observed values ('evidence stays in-env') and body excerpts are capped; PARTIAL: the tester's option to cut excerpts was not found.
    C-076OneDroid Synapse is the governed MCP gateway.onedroid.ai/llms-full.txt (sources/llms-full.txt L227 (PAGE /; under "It brings its own run layer."))https://synapse.onedroid.ai/#200 text/html, <title>OneDroid MCP Router</title>
    The Synapse host answers 200 with the title 'OneDroid MCP Router'; PARTIAL: the 'governed' part is not shown by the public page.
    C-077judges the result the way an operator would: by reading the system's own structured logs, its saga events, and its database state, from a process that sits next to itdocs.onedroid.ai/argus (sources/argus-docs.html L63-65 (under "Argus — automated end-to-end testing for AI-written code"))
    docs.onedroid.ai/argus.md (sources/argus-docs.md L10-12)
    repo/internal/mcpserver/tools.go#L65-71
    repo/internal/obsquery/obsquery.go#L1-3
    repo/internal/scenario/validate.go#L474
    The runner reads saga events (runner_get_sagas), Loki logs and read-only DB rows from inside the environment.
    C-079The agent skills and the MCP server are released.onedroid.ai/llms.txt (sources/llms.txt L45 (under "OneDroid"))repo/skills/scenario-author/SKILL.md
    repo/internal/mcpserver/tools.go#L34-77
    repo/LICENSE#L1
    The public MIT repo ships an agent skill (skills/scenario-author) and an MCP server with registered tools.
    C-080an MCP server with separate credentials for the tester and the builderonedroid.ai/llms-full.txt (sources/llms-full.txt L2916 (PAGE /argus; under "Three ways in."))
    onedroid.ai/argus.md (sources/argus.md L180 (under "Three ways in."))
    repo/internal/auth/auth.go#L1-13
    repo/internal/auth/auth.go#L35-38
    The MCP server maps two distinct tokens to two roles (product runner_* only, test runner_* + author_*) and refuses to start if they are equal.
    C-081prefixed `odts_`, shown once [...] a **builder token** (**Settings → API Tokens**, then **Generate builder token**), which reaches only the control plane's `runner_*` toolsdocs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L94-134 (under "2. The two kinds of token"))repo/internal/doctor/token.go#L19-20
    repo/internal/auth/auth.go#L8
    repo/internal/mcpserver/tools.go#L34-77
    odts_ tokens are recognised as author PATs and the builder hat reaches only runner_* tools; PARTIAL: 'shown once' and the Settings path are in the closed web UI.
    C-086The site no longer describes things that are not built: the one-line install command and the CI action are gone from it.onedroid.ai/llms-full.txt (sources/llms-full.txt L6158-6160 (PAGE /changelog; under "The site now leads with what testing costs"))
    onedroid.ai/changelog.md (sources/changelog.md L49-51 (under "The site now leads with what testing costs"))
    sources/changelog.md#L49-51
    sources/ (grep 'curl ... | sh', 'install.sh', 'GitHub Action', 'uses: OneDro1d' over argus.md, argus-docs.md, argus-quickstart.md, pricing.md, llms.txt, llms-full.txt outside the L320-2735 register block: no Argus one-line install or CI action)
    The saved current pages carry no one-line Argus install command and no CI action (the only install.sh hit is the separate /kit installer), so the removal is checkable in the corpus.
    C-087A run with faults gets certificate format `argus-certificate/v4`.docs.onedroid.ai/argus.md (sources/argus-docs.md L169 (under "What changed recently"))repo/internal/certificate/certificate.go#L31-35
    FormatV4 'argus-certificate/v4' is issued only for a run that applied at least one fault; PARTIAL: a run with faults that also has preconditions gets v5 (L37-40).
    C-088The runner and the CLI are open source under MITonedroid.ai/llms-full.txt (sources/llms-full.txt L2939 (PAGE /argus; under "MCP server"))
    onedroid.ai/argus.md (sources/argus.md L203 (under "MCP server"))
    repo/LICENSE#L1
    repo/README.md#L3-4
    repo/README.md#L140
    The MIT LICENSE covers the repository that holds both the runner and the argus CLI (cmd/argus).
    C-089The AMQP sampler under `jmeter-plugins/amqp` keeps its own Apache-2.0 licence file.raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L12 (under "OneDroid Argus runner"))repo/jmeter-plugins/amqp/LICENSE#L1-2
    jmeter-plugins/amqp/LICENSE is the Apache License, Version 2.0.
    C-090No fee, no seat count, no expiryonedroid.ai/llms-full.txt (sources/llms-full.txt L5920 (PAGE /pricing; under "The OneDroid Argus runner"))
    onedroid.ai/pricing.md (sources/pricing.md L35 (under "The OneDroid Argus runner"))
    repo/LICENSE#L5
    The MIT licence grants permission 'free of charge' with no seat or term limit.
    C-091The container image is published at `ghcr.io/onedro1d/argus-runner`.onedroid.ai/llms-full.txt (sources/llms-full.txt L6255-6256 (PAGE /changelog; under "The OneDroid Argus runner is open source"))
    onedroid.ai/changelog.md (sources/changelog.md L146-147 (under "The OneDroid Argus runner is open source"))
    https://github.com/OneDro1d/argus-runner/pkgs/container/argus-runner#200 <title>Package argus-runner · GitHub</title>, 'docker pull ghcr.io/onedro1d/argus-runner:sha-1c9822f'
    repo/README.md#L50
    The public GitHub package page for ghcr.io/onedro1d/argus-runner answers 200 with a pull command.
    C-092Not in it: the hosted control plane (registry, signing, anchoring, the web UI).onedroid.ai/llms-full.txt (sources/llms-full.txt L6258 (PAGE /changelog; under "The OneDroid Argus runner is open source"))
    onedroid.ai/changelog.md (sources/changelog.md L149 (under "The OneDroid Argus runner is open source"))
    repo/README.md#L9-10
    repo/README.md#L150-156
    The public repo has no control-plane, signing, anchoring or web-UI code ('argus serve --mode control' is refused in this build), which matches the changelog's list.
    C-094the onboarding kit they drive (`onboarding/onboard.sh` and the compose and k3d scripts) is not shipped hereraw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L99-100 (under "Quickstart"))repo/onboarding/ (listing: HOW-TO-ARGUS-CONFIG.md, argus-config.template.yaml only)
    repo/README.md#L99-101
    onboarding/ in the clone at ba724fb1 holds no onboard.sh and no compose or k3d scripts, so the kit is indeed not shipped.
    C-095No token, no private module and no database are needed.raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L160-161 (under "Build and test"))repo/go.mod#L1-L50 (module github.com/OneDro1d/argus-runner; every require is a public module: go-ethereum, google/uuid, hamba/avro, jackc/pgx, rabbitmq/amqp091-go, yaml.v3 and public indirects)
    repo/BUILDING.md#L7 'No GitHub token, no private module, no database, no Docker.'
    repo/BUILDING.md#L41 'The suite needs no network and no database.'
    go.mod at ba724fb1 names only public modules, so 'no private module' is backed by the manifest; PARTIAL: 'no token' and 'no database' rest on BUILDING.md's own statement, since the build and test suite were not run (the bound forbids running repo code).
    C-101"private":false [...] "created_at":"2026-10-07T22:26:48Z" [...] "pushed_at":"2026-10-27T17:40:27Z" [...] "stargazers_count":1 [...] "forks_count":0 [...] "archived":false [...] "open_issues_count":0 [...] "spdx_id":"MIT" [...] "visibility":"public"api.github.com/repos/OneDro1d/argus-runner (sources/github-repo.json L1)https://api.github.com/repos/OneDro1d/argus-runner#HTTP 200 'private':false 'created_at':'2026-10-07T22:26:48Z' 'pushed_at':'2026-10-27T17:40:27Z' 'stargazers_count':1 'forks_count':0 'archived':false 'open_issues_count':0 'spdx_id':'MIT' 'visibility':'public'
    sources/github-repo.json#L1
    repo/LICENSE#L1 'MIT License'
    A live GET of the GitHub API returns every quoted field with the quoted value, identical to the saved sources/github-repo.json, and the repo's LICENSE is MIT.
    C-1022026-10-09 ### The OneDroid Argus runner is open sourceonedroid.ai/llms-full.txt (sources/llms-full.txt L6247-6249 (PAGE /changelog; under "OneDroid Argus 0.3.70: a check can read the database and the logs, and make the system fail"))
    onedroid.ai/changelog.md (sources/changelog.md L138-140 (under "OneDroid Argus 0.3.70: a check can read the database and the logs, and make the system fail"))
    https://api.github.com/repos/OneDro1d/argus-runner#HTTP 200 'visibility':'public' 'spdx_id':'MIT' 'created_at':'2026-10-07T22:26:48Z'
    repo/LICENSE#L1 'MIT License'
    sources/changelog.md#L138-140
    The runner repository is public under MIT and was created 7 Oct, before the 9 Oct entry; PARTIAL: the date it was made public is not exposed by the API, so the 2026-10-09 date itself is not evidenced.
    C-113OneDroid Synapse speaks MCP over HTTP, so any MCP-compatible client works, with any model provider. Connecting Claude Code takes a single claude mcp add command with a bearer token.onedroid.ai/llms-full.txt (sources/llms-full.txt L4176 (PAGE /synapse; under "Does OneDroid Synapse work with Claude Code and Cursor?"))https://synapse.onedroid.ai/agent/mcp#HTTP 401 application/json 'missing_token'; Access-Control-Expose-Headers: WWW-Authenticate, Mcp-Session-Id; WWW-Authenticate: Bearer ... oauth-protected-resource
    sources/llms-full.txt#L4176
    The live endpoint answers as an HTTP MCP server with bearer auth (MCP session headers, Bearer challenge); PARTIAL: 'any MCP client, any model provider' and the single claude mcp add command were not exercised (no token).
    C-114MCP endpoint for token clients: `https://synapse.onedroid.ai/agent/mcp` [...] OAuth clients use `https://synapse.onedroid.ai/hub/<slug>/mcp` instead, where the path selects the hub.onedroid.ai/llms.txt (sources/llms.txt L120-124 (under "Connecting an agent"))https://synapse.onedroid.ai/agent/mcp#HTTP 401 'missing_token' (bearer endpoint exists)
    sources/llms.txt#L120-124
    The token-client MCP endpoint exists and demands a bearer token; PARTIAL: the OAuth hub path /hub/<slug>/mcp is a placeholder and was not exercised.
    C-115A good token returns HTTP 200 with a JSON-RPC result; a wrong one returns `401 invalid_token`; a missing header returns `401 missing_token`.onedroid.ai/llms-full.txt (sources/llms-full.txt L4445-4447 (PAGE /quickstart; under "The check"))https://synapse.onedroid.ai/agent/mcp#HTTP 401 {'code':'missing_token','message':'missing authentication token'} (GET, no Authorization header)
    sources/llms-full.txt#L4445-4447
    A request with no header returns 401 missing_token exactly as documented; PARTIAL: the good-token 200 and the wrong-token 401 invalid_token were not exercised (no credentials sent).
    C-116Send the same `initialize` probe as above to `https://engram.onedroid.ai/mcp` with your Engram token. A good token gets HTTP 200 and a server named `engram`; no token, a wrong one or a revoked one gets HTTP 401 with a plain-text `Unauthorized`.onedroid.ai/llms-full.txt (sources/llms-full.txt L4507-4510 (PAGE /quickstart; under "The check"))https://engram.onedroid.ai/mcp#HTTP 401 text/plain 'Unauthorized', WWW-Authenticate: Bearer realm='engram' (GET, no token)
    sources/llms-full.txt#L4507-4510
    With no token the Engram MCP endpoint returns 401 plain-text Unauthorized as documented (the harness's 404 came from a trailing backtick in the URL); PARTIAL: the good-token 200 with server name engram and the wrong/revoked-token paths were not exercised.
    C-137OneDroid Synapse implements OAuth 2.0 authorization-server discovery at `/.well-known/oauth-authorization-server`, dynamic client registration, and the device authorization flowonedroid.ai/llms-full.txt (sources/llms-full.txt L4724-4726 (PAGE /security; under "Standards, not a bespoke scheme."))
    onedroid.ai/security.md (sources/security.md L115-117 (under "Standards, not a bespoke scheme."))
    https://synapse.onedroid.ai/.well-known/oauth-authorization-server#HTTP 200 'issuer':'https://synapse.onedroid.ai' 'registration_endpoint':'https://synapse.onedroid.ai/oauth/register' 'device_authorization_endpoint':'https://synapse.onedroid.ai/oauth/device' 'grant_types_supported' includes 'urn:ietf:params:oauth:grant-type:device_code'
    sources/security.md#L115-117
    The live RFC 8414 discovery document at the stated path lists a dynamic client registration endpoint and a device authorization endpoint with the device_code grant (the harness's no-host error came from the bare path).
    C-138User identity is handled by Clerk, so OneDroid Synapse inherits its enterprise SSO support: SAML 2.0 and OIDC against your own identity provider, including Microsoft Entra ID, Okta and Google Workspace.onedroid.ai/llms-full.txt (sources/llms-full.txt L4733-4736 (PAGE /security; under "Standards, not a bespoke scheme."))
    onedroid.ai/security.md (sources/security.md L124-127 (under "Standards, not a bespoke scheme."))
    https://clerk.onedroid.ai/v1/environment#fetched/008.txt L1 (HTTP 200, 'application_name':'OneDroid', 'instance_environment_type':'production')
    fetched/008.txt#L1 'saml':{'enabled':false},'enterprise_sso':{'enabled':false}
    sources/security.md#L124-127
    A production Clerk instance named OneDroid serves the identity layer, so 'identity is handled by Clerk' is backed; PARTIAL: SAML 2.0/OIDC enterprise SSO is not enabled on that instance (saml and enterprise_sso enabled:false), so the SSO clause is a plan capability with no artefact.
    C-145Our website sets no advertising or analytics cookies. [...] We also forward the request logs of this website and of docs.onedroid.ai (requested page, time, IP address, user agent, referrer) to a log store we run ourselves in the EU, where they are kept for 30 daysonedroid.ai/llms-full.txt (sources/llms-full.txt L6457-6464 (PAGE /privacy; under "What we collect"))
    onedroid.ai/privacy.md (sources/privacy.md L66-73 (under "What we collect"))
    https://onedroid.ai/#HTTP 200, server: Vercel, no set-cookie header in the response
    https://onedroid.ai/#HTML has no googletagmanager/gtag/google-analytics/plausible/segment/hotjar/mixpanel/posthog/_vercel/insights/clarity script (the only grep hit is quoted text inside the embedded claims register)
    sources/privacy.md#L66-73
    The home page sets no cookie in its response and loads no advertising or analytics script, which would differ if it did; PARTIAL: forwarding request logs of onedroid.ai and docs.onedroid.ai to an EU log store kept 30 days is server-side and not observable.
    C-147Amazon Web Services and Microsoft Azure (application hosting), Supabase (platform database hosting), Clerk (authentication), Vercel (website hosting), OpenAI (computing search embeddings for content you store in OneDroid Engram, under terms that prohibit training on your data), and Google Workspace (support email).onedroid.ai/llms-full.txt (sources/llms-full.txt L6525-6529 (PAGE /privacy; under "How we share data"))
    onedroid.ai/privacy.md (sources/privacy.md L134-138 (under "How we share data"))
    https://onedroid.ai/#HTTP 200 'server: Vercel', 'x-vercel-id' (website hosting)
    fetched/008.txt#L1 (clerk.onedroid.ai/v1/environment HTTP 200, a Clerk production instance: authentication)
    sources/privacy.md#L134-138
    Two of the named subprocessors are observable (Vercel serves the website, Clerk serves authentication); PARTIAL: AWS/Azure hosting, Supabase, OpenAI embeddings under no-training terms and Google Workspace support email are not shown by any public artefact.
    C-151our total liability for all claims relating to the Service is limited to the greater of USD 100 or the amounts you paid us for the Service in the twelve months before the claim aroseonedroid.ai/llms-full.txt (sources/llms-full.txt L6742-6744 (PAGE /terms; under "10. Limitation of liability"))https://onedroid.ai/terms.md#HTTP 200 text/markdown, L158 'greater of USD 100 or the amounts you paid us for the Service in the twelve months'
    sources/llms-full.txt#L6742-6744
    The liability cap is a contractual term, and the live Terms of Service, the instrument that creates it, contain the quoted clause.
    C-155The Dark Factory kits are open source under Apache-2.0onedroid.ai/llms-full.txt (sources/llms-full.txt L265 (PAGE /; under "Open where it has to be."))https://api.github.com/repos/OneDro1d/dark-factory#HTTP 200 'private':false 'visibility':'public' 'license.spdx_id':'Apache-2.0'
    https://api.github.com/repos/OneDro1d/dark-factory/git/trees/HEAD?recursive=1#HTTP 200 path 'LICENSE'
    sources/llms-full.txt#L265
    The dark-factory repository at github.com/OneDro1d is public and GitHub detects its LICENSE as Apache-2.0.
    C-156Seven kits today: `method-core` (the method itself), `dev`, `knowledge-worker`, `agent-ops`, `code-review`, `distributed-systems`, `frontend`.onedroid.ai/llms-full.txt (sources/llms-full.txt L4546-4549 (PAGE /quickstart; under "Clone and list the kits"))https://api.github.com/repos/OneDro1d/dark-factory/git/trees/HEAD?recursive=1#HTTP 200 paths kits/method-core, kits/dev, kits/knowledge-worker, kits/agent-ops, kits/code-review, kits/distributed-systems, kits/frontend
    sources/llms-full.txt#L4546-4549
    The public repo tree has exactly the seven named kit directories under kits/ (repo-grep missed them because D/repo is the Argus runner, not dark-factory).
    C-157It ships with a self-test that plants a known-positive canary for every scan class and asserts the gate fails on eachonedroid.ai/llms-full.txt (sources/llms-full.txt L6282-6283 (PAGE /changelog; under "The Dark Factory method is now open source"))
    onedroid.ai/changelog.md (sources/changelog.md L173-174 (under "The Dark Factory method is now open source"))
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/boot-kit/scripts/gate-selftest.sh#HTTP 200 L13 'This script plants a known-positive canary for every'
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/boot-kit/scripts/gate-selftest.sh#L142 'every pattern class must FAIL on every canary'
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/boot-kit/scripts/gate-selftest.sh#L149-161 (per-class loop writes each canary and runs the gate)
    The public gate self-test script loops over every pattern class, plants each class's canaries and asserts the gate fails on each, and errors if a class has no canary (L152).
    C-158The runbook ends with a validation run that commits its own report into your copy.onedroid.ai/llms-full.txt (sources/llms-full.txt L4578-4579 (PAGE /quickstart; under "The check"))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/boot-kit/scripts/validate.sh#HTTP 200 L241-245 (copies REPORT.md to VALIDATE-REPORT-<stamp>-<instance>.md at the kit root)
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/boot-kit/scripts/validate.sh#L262-292 ('commit + push the report', commit message 'M-VALIDATE: validation report')
    sources/llms-full.txt#L4578-4579
    The public validate.sh copies the validation report into the copy and git-commits it by path; PARTIAL: that the runbook (START-HERE.md) ends with this run was not traced.
    C-167Every onedroid.ai page above except the blog (which has its own index at https://onedroid.ai/blog/llms.txt) and the third-party evidence pages, concatenated into one document, generated from the HTML.onedroid.ai/llms.txt (sources/llms.txt L129 (under "Optional"))https://onedroid.ai/llms-full.txt#HTTP 200 sha256 2b063c3ac9cf2b36…
    sources/llms-full.txt#L16-L6597 ('# PAGE:' sections for /, /argus, /argus-agents, case studies, /synapse … /terms; no /blog or /evidence PAGE section)
    https://onedroid.ai/blog/llms.txt#HTTP 200 text/plain
    llms-full.txt is served and is one document of '# PAGE:' sections for the site pages with no blog post or third-party evidence page among them, and the blog's own index answers 200; 'generated from the HTML' is not separately shown.
    C-168Every post is served as markdown at `<url>.md`; machine index at https://onedroid.ai/blog/llms.txt, Atom feed at https://onedroid.ai/blog/feed.xml.onedroid.ai/llms.txt (sources/llms.txt L74 (under "Blog"))https://onedroid.ai/blog/llms.txt#HTTP 200 text/plain (lists 10 posts as <url>.md)
    https://onedroid.ai/blog/feed.xml#HTTP 200 application/atom+xml
    https://onedroid.ai/blog/how-to-check-a-merged-fix-is-live.md#HTTP 200 text/markdown
    The machine index and Atom feed answer 200 with the right content types and a sampled post is served as markdown at <url>.md; PARTIAL: only one post's .md twin was checked, not every post.
    C-171The limits are part of the result, so they are in the text and not in a footnote.onedroid.ai/llms-full.txt (sources/llms-full.txt L3125 (PAGE /case-studies; under "What happened when"))sources/case-study-otrs-znuny.md#L149 '## What this test did not cover'
    sources/case-study-finos-calm.md#L143 '## What this test did not cover'
    sources/case-study-claude-code-alone-vs-tester.md#L99 '## What this experiment did not cover'
    Each of the five saved case studies carries its limits as an in-text section heading, not a footnote.
    C-179If the cluster does not enforce NetworkPolicy, the dependent checks end `not_applicable` instead of `failed`, and a run where every check is `not_applicable` has the verdict `not_applicable`.docs.onedroid.ai/argus.md (sources/argus-docs.md L117-119 (under "What changed recently"))repo/internal/argus/preconditions.go#L19 'absent -> NOT executed: status not_applicable'
    repo/docs/preconditions.md#L8 (cluster that does not enforce NetworkPolicy)
    repo/docs/preconditions.md#L34-35 (networkpolicy.ingress_enforced / egress_enforced probes)
    The open runner implements NetworkPolicy precondition probes whose absent outcome marks dependent checks not_applicable, and the verdict table makes an all-not_applicable run not_applicable.
    C-180`targets.auth` takes `type: clerk_session`, and a named http target can carry its own `auth:`. The executor mints a short-lived session per run.docs.onedroid.ai/argus.md (sources/argus-docs.md L123-124 (under "What changed recently"))repo/internal/config/clerksession.go#L13 'A SUT behind Clerk accepts a short-lived session JWT ... The executor mints one'
    repo/internal/config/clerksession.go#L66 (a named target's own auth: wins, else targets.auth)
    repo/onboarding/HOW-TO-ARGUS-CONFIG.md#L53 ('type: clerk_session ... mints a short-lived session token per run'; a named http_targets.<name> may carry its own auth:)
    The config code and schema implement targets.auth type clerk_session with per-run minting and a per-named-target auth: override.
    C-181the author's report no longer carries a value your system returneddocs.onedroid.ai/argus (sources/argus-docs.html L177-178 (under "What changed recently"))
    docs.onedroid.ai/argus.md (sources/argus-docs.md L126-127 (under "What changed recently"))
    repo/internal/chain/mcpgateway.go#L302-304 ('EVERY claim is shown AS WRITTEN (${saved.x})' ... the bound copy 'put a value the SUT returned into the author's report')
    repo/internal/chain/body_excerpt.go#L14-15 (credential and saved-value scrub before the cut)
    sources/argus-docs.md#L126-127
    The open runner writes mcp-step claims as written rather than bound to returned values; PARTIAL: the author's report is assembled by the closed control plane, so the end-to-end absence there is not shown.
    C-182on an instance that declares `money_handling`, `money_writes.allow` can allow a runnable `## CLEANUP` by form, a `ui` spec, and `fault` steps. A refused scenario does not run its `## CLEANUP`.docs.onedroid.ai/argus.md (sources/argus-docs.md L128-130 (under "What changed recently"))repo/internal/config/moneywrites.go#L43-47 (allow entries engine: cleanup with form, engine: ui with spec, engine: fault)
    repo/internal/config/moneywrites.go#L103-110 (money_writes refused unless money_handling is true)
    repo/internal/argus/cleanup.go#L122-137 (refusedCleanup: 'the scenario was refused before firing, so this ... cleanup block was NOT executed')
    The open config code accepts cleanup-by-form, ui-spec and fault allow entries only beside money_handling, and a refused scenario's cleanup is recorded as not executed.
    C-183an `http` step whose response body is over 1 MiB reads `error` when it has a body claim or a `save`, and the chain's deadline now adds each polled step's `poll.timeout`.docs.onedroid.ai/argus.md (sources/argus-docs.md L132-133 (under "What changed recently"))repo/internal/chain/http.go#L44
    repo/internal/chain/http.go#L405
    repo/internal/argus/chain_scenario.go#L572
    The open runner caps an http step body at 1 MiB (maxHTTPBodyBytes) and treats an over-limit body specially only when the step has body claims or a save, and the chain backstop adds each step's poll.timeout (stepPollTimeouts).
    C-184`body has total == 42`, `within 2% of`, `body not contains`, `body has no <field>`, `body every <list> has …`, and `elapsed < 500ms` on an `http` step.docs.onedroid.ai/argus.md (sources/argus-docs.md L136-138 (under "What changed recently"))repo/skills/scenario-author/SKILL.md#L1011-1017
    repo/internal/scenario/bodyassert.go#L110
    repo/internal/scenario/bodyassert.go#L116
    Every listed claim form (== number, within N% of, body not contains, body has no field, body every list has, elapsed < bound) is parsed in internal/scenario and documented in the open skill's claim table.
    C-185a run that is going shows the counts so far and `planned`; a step can save a JSON value and send it in a later step; a chain `http` step under a non-JSON `Content-Type` sends a string body as text; `## TIMEOUT` governs a browser check, up to 300 s.docs.onedroid.ai/argus.md (sources/argus-docs.md L140-143 (under "What changed recently"))repo/internal/argus/tally.go#L58-83
    repo/internal/scenario/savedjson.go#L12
    repo/internal/scenario/textbody.go#L16-49
    The tally carries a planned count beside running counts, a save can take the json form, TextBodyOf sends a string body as text under the step's own non-JSON Content-Type, and a ui scenario's TIMEOUT ceiling is 300s in code.
    C-186read the message a request caused without taking it from the system's own consumer.docs.onedroid.ai/argus (sources/argus-docs.html L208-209 (under "What changed recently"))
    docs.onedroid.ai/argus.md (sources/argus-docs.md L152-153 (under "What changed recently"))
    repo/docs/COMPATIBLE-SUT.md#L176-180
    repo/docs/COMPATIBLE-SUT.md#L258
    sources/argus-docs.md#L152-153
    The open runner's SUT contract documents a private tap queue the runner declares, binds and deletes, which 'never consumes from YOUR queues'; PARTIAL: the tap is described in docs, the code path was not traced line by line.
    C-187on an instance that declares `money_handling`, the owner's `money_writes.allow` list can name an HTTP route, an amqp operation, an `mcp` tool, a `db` step or a `logs` step.docs.onedroid.ai/argus.md (sources/argus-docs.md L155-157 (under "What changed recently"))repo/internal/config/config.go#L331
    repo/internal/config/moneywrites.go#L40-64
    repo/internal/chain/http.go#L164
    money_handling is a config key and a money_writes.allow entry takes an engine of http (default), amqp, mcp, db or logs (plus cleanup, ui, fault) in the open config code.
    C-188**An `mcp` step can read data a gateway wraps in text** (v0.3.70): `extract`, claims written `body extracted …`, `unavailable` and `poll`.docs.onedroid.ai/argus.md (sources/argus-docs.md L159-160 (under "What changed recently"))repo/internal/chain/mcpgateway.go#L21-38
    repo/internal/chain/mcpgateway.go#L67-81
    repo/internal/chain/mcpgateway.go#L375-377
    The mcp gateway step implements extract (regex and path), claims written body extracted, an unavailable pattern and poll; PARTIAL: the version label v0.3.70 is not tied to any tag in the clone (no tags present).
    C-189`argus render-k8s --collect-sut-logs --sut-logs-via api`, for a cluster that refuses host mounts.docs.onedroid.ai/argus.md (sources/argus-docs.md L163 (under "What changed recently"))repo/cmd/argus/main.go#L270
    repo/cmd/argus/sut_logs_via.go#L31-34
    render-k8s registers --sut-logs-via node|api, where api renders a one-replica Deployment reading the Kubernetes API with no hostPath, which is the host-mount-free mode the claim names.
    C-190`argus cloud-enroll --workspace <id>` and `argus cloud-create-workspace --name <name>`.docs.onedroid.ai/argus.md (sources/argus-docs.md L173 (under "What changed recently"))repo/cmd/argus/main.go#L229
    repo/cmd/argus/main.go#L2223
    repo/cmd/argus/usage_commands.go#L26-28
    cloud-enroll takes --workspace and cloud-create-workspace requires --name in the open CLI's flag code.
    C-191a new `router-wiring` check, `--tester` runs the cluster checks without a tester.envdocs.onedroid.ai/argus.md (sources/argus-docs.md L175-176 (under "What changed recently"))repo/internal/doctor/router.go#L69
    repo/cmd/argus/doctor_tester.go#L59-62
    A doctor check with ID router-wiring exists, and doctor --tester runs its kubectl phases when the tester token phase fails (a missing tester.env no longer skips every phase).
    C-192**A target that runs in no cluster** (v0.3.63): `outside_cluster: true`.docs.onedroid.ai/argus.md (sources/argus-docs.md L204 (under "What changed recently"))repo/internal/argus/environment.go#L94
    repo/internal/argus/environment.go#L147
    repo/onboarding/HOW-TO-ARGUS-CONFIG.md#L202
    outside_cluster: true on a test target is read by the environment capture and marks Kubernetes reads not applicable; PARTIAL: the version label v0.3.63 is not tied to a tag in the clone.
    C-193`- step <name>: unreachable` passes only when the connection could not be made, and `argus calm import` turns a CALM architecture into checks.docs.onedroid.ai/argus.md (sources/argus-docs.md L191-192 (under "What changed recently"))repo/internal/chain/claim_unreachable.go#L18-20
    repo/internal/chain/claim_unreachable.go#L136
    repo/internal/calm/import.go#L829
    The unreachable claim passes only on a transport failure (a connection that succeeds fails the step), and argus calm import is implemented in internal/calm.
    C-195a tester can **certify** a build: commit to what is being certified, write and seal a set of tests the builder never sees, then run them once as a `final` run. Argus anchors each step on a blockchain as it happens: the sealed set before the run, the verdict after it, and the moment the tests are revealed.docs.onedroid.ai/argus.md (sources/argus-docs.md L105-108 (under "Certify, and let anyone check it"))repo/docs/CERTIFICATE.md#L10-22
    repo/internal/certificate/verify.go#L300
    https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#200 block 47817413 2026-10-07T19:58:34Z
    The certificate format documents set (v2), verdict (v3) and reveal anchors of a final run, and the FINOS case's public Base Sepolia records show a seal tx before the verdict tx; PARTIAL: the anchoring writes happen in the closed control plane and the reveal anchor was not observed on chain.
    C-196A scenario is a `TRIGGER` (what to do — an HTTP call, a message, a database read, an MCP tool call), a `VERIFY` (where to look), and an `EXPECT` (what "healthy" means there).docs.onedroid.ai/argus.md (sources/argus-docs.md L73-74 (under "What gets checked"))repo/examples/http-hello/scenarios/HELLO-001-health-200.md#L8
    repo/examples/http-hello/scenarios/HELLO-001-health-200.md#L11
    repo/examples/http-hello/scenarios/HELLO-001-health-200.md#L14
    The example scenario in the open repo is built of TRIGGER, VERIFY and EXPECT sections, and the parser reads trigger/verify/expect (internal/scenario/parse.go L512).
    C-197This holdout is enforced twice, not once: by what a builder's credential is even allowed to call, and separately by what its own environment is allowed to hold a copy of.docs.onedroid.ai/argus (sources/argus-docs.html L109-110 (under "The holdout"))
    docs.onedroid.ai/argus.md (sources/argus-docs.md L62-63 (under "The holdout"))
    repo/internal/role/role.go#L1-20
    repo/internal/onboard/guard.go#L14-51
    role.go gates scenario access by the bearer token's hat, and FolderGuard separately refuses a product folder that contains or nests the scenarios, which are the two enforcement points the claim names.
    C-200and a Kubernetes cluster that can reach both your system and the control plane (the runner connects out only; it needs no inbound route)raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L43-44 (under "Quickstart"))repo/internal/federation/wire.go#L12
    repo/internal/federation/federation.go#L2
    sources/argus-runner-README.md#L43-44
    The executor's federation client is implemented as outbound-only HTTPS to the control plane, with no inbound connection to the in-env runner.
    C-201Runs are requested through the control plane's author tools (for example from an agent connected to its MCP endpoint, `https://argus-dev.onedroid.ai/mcp`).raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L95-96 (under "Quickstart"))repo/internal/sessioninit/sessioninit.go#L217
    repo/cmd/argus/certificate_cmd.go#L93
    https://argus-dev.onedroid.ai/mcp#404 to a GET (MCP is POST)
    The open CLI wires an agent's MCP config and its own author calls to <control-plane>/mcp; PARTIAL: a plain GET of argus-dev.onedroid.ai/mcp answers 404, so the live author-tools endpoint was not observed without a session.
    C-202`--rpc` is your own chain client; `--chains` is your own copy of the chains' allowed signers (without it, each anchor is only checked against the signer it declares); `--btc-headers` is a block-explorer URL or a local file for OpenTimestamps receipts.raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L110-112 (under "Verify a certificate"))repo/cmd/argus/certificate_cmd.go#L14
    repo/docs/CERTIFICATE.md#L186-195
    repo/docs/CERTIFICATE.md#L278-284
    certificate verify registers --rpc, --chains and --btc-headers, and the docs and verify code describe each as the reader's own chain client, allowed-signers file and block-explorer or file header source.
    C-203`argus certificate verify` re-reads those anchors from a chain client you choose and checks that the certificate's declared fields (verdict, tallies, set hash, artifact digest, time) are exactly what was anchored, and that each anchor was written by a signer you allow. A certificate edited after the fact fails this check.raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L20-23 (under "What a OneDroid Argus certificate proves today, and what it does not"))repo/internal/certificate/verify.go
    repo/docs/CERTIFICATE.md#L39-44
    repo/docs/CERTIFICATE.md#L282-297
    verify.go reads each anchor back through the chosen client and compares the declared set and verdict fields to the anchored payload under the signer check, so an edited certificate reads MISMATCH.
    C-204| `cmd/argus` | the CLI: [...] `internal/certificate`, `internal/chainread` | the certificate format and the read-only chain reader behind `certificate verify` [...] `templates/` | the JMeter templates the runner drives [...] `jmeter-plugins/amqp` | the AMQP JMeter sampler (Java) [...] `testkit/ui` | the Playwright harness for Web UI scenarios [...] `examples/http-hello` | a minimal kitraw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L140-148 (under "What is here"))repo/cmd/argus
    repo/internal/certificate
    repo/internal/chainread
    Every directory the README lists exists in the clone with the stated content (JMeter templates, the AMQP sampler, a Playwright harness, a minimal kit).
    C-205The control plane and its store, the web UI, anything that writes to a chain or holds a chain key, reveal document assembly (checking a reveal and replaying it ARE here, see above), deployment overlays and secrets for any hosted environment, and the onboarding kit. [...] `argus serve --mode control` is refused in this build.raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L152-156 (under "What is deliberately not here"))repo/cmd/argus/main.go#L2905-2906
    repo/README.md#L152-156
    serve --mode control is refused in code with 'not part of this build', and the clone holds no control-plane, web UI or chain-writing component; PARTIAL: absence of each listed item was judged from the tree, not exhaustively.
    C-206`--scope` is required from v0.3.41 on, and there is no default: `author` for a tester or operator (the `cloud-*` commands and the test-agent tools). `runner` asks for builder tools only, but a builder does not need itdocs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L166-168 (under "3. Point the CLI at your control plane"))repo/cmd/argus/main.go#L215-219
    repo/cmd/argus/main.go#L1494-1496
    cloud-login's --scope has no default and refuses an empty value, naming author (operator/tester tools) and runner (builder tools); PARTIAL: 'from v0.3.41' is not tied to a tag in the clone.
    C-207Local checks and setup commands run without a token (`preflight` reports a missing control-plane token as a blocking finding and still runs), for example `validate-scenario`, `validate-config`, `package-check`, `propose-from-repo`, `calm import`, `preflight`, `doctor` and `argus version`docs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L107-109 (under "2. The two kinds of token"))repo/cmd/argus/main.go#L975
    repo/cmd/argus/main.go#L560-612
    repo/internal/preflight/preflight.go#L85-89
    The CLI's auth text exempts validate-config, validate-scenario, package-check and propose-from-repo from the token, doctor and render steps run pre-auth, and preflight models a missing token as a finding; PARTIAL: calm import and version were not traced to the pre-auth path.
    C-208That hat reaches the same six-tool runner surface the author hat also has (`validate-config`, `run`, `get-report`, `get-sagas`, `tail-logs`, `get-dashboard-url`). Neither path reaches the scenario-authoring commandsdocs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L139-141 (under "2. The two kinds of token"))repo/internal/mcpserver/tools.go#L34-77
    repo/internal/role/role.go#L17-20
    The runner MCP surface registers exactly six runner tools (validate_config, run, get_report, get_sagas, get_tail_logs, get_dashboard_url) and the product hat has no scenario access.
    C-209On a tester machine, use `argus doctor --tester` instead (v0.3.47 or later). It prints one line per onboarding phase, each PASS, WARN, FAIL or SKIP [...] It exits `4` on any FAIL; a WARN does not change the exit code. [...] from v0.3.52 that check is a warning ("not checked") and `doctor` still exits `0`docs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L236-237, L254-255, L259 (under "4. Prove read access before wiring anything else"))repo/cmd/argus/doctor_tester.go#L20
    repo/cmd/argus/doctor_tester.go#L413
    repo/cmd/argus/main.go#L56
    doctor --tester prints one PASS/WARN/FAIL/SKIP line per phase and returns exitFailed, which is 4; PARTIAL: the v0.3.47 and v0.3.52 version boundaries are not tied to tags in the clone.
    C-210`ARGUS_EXECUTOR_SECRET` was called `ARGUS_AUTHOR_TOKEN`, and `ARGUS_CP_AUTHOR_TOKEN` was called `ARGUS_CP_TOKEN`. The old names were easy to mistake for a person's author token. CLI releases after v0.3.39 read the new names and still accept the old ones, with a warning.docs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L101-104 (under "2. The two kinds of token"))repo/internal/envname/envname.go#L4-7
    repo/internal/envname/envname.go#L31-37
    repo/internal/envname/envname.go#L54-119
    envname maps ARGUS_AUTHOR_TOKEN to ARGUS_EXECUTOR_SECRET and ARGUS_CP_TOKEN to ARGUS_CP_AUTHOR_TOKEN, still reads the old names and warns once; PARTIAL: 'after v0.3.39' is not tied to a tag.
    C-212`--tier` is `aks`, `k3d` or `managed`: `managed` for any other cluster, including one you run yourself (k3s, kubeadm). `kind` and `minikube` are not tiers; use `k3d` for them, which renders identically. `render-k8s` also accepts `eks` and `gke`.docs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L203-205 (under "4. Prove read access before wiring anything else"))repo/cmd/argus/main.go#L238
    repo/cmd/argus/main.go#L3446
    repo/docs/COMPATIBLE-SUT.md#L52-57
    render-k8s's --tier flag lists k3d, aks, eks, gke and managed (managed for a self-run cluster such as k3s) and the compatibility doc and code treat kind and minikube as k3d.
    C-213The Argus Code runner runs inside your environment, and its source is open under MIT. No credit card.onedroid.ai/llms-full.txt (sources/llms-full.txt L44 (PAGE /; under "Test the agents where they work: in the workflow, in the fleet, next to the people who sign off."))repo/LICENSE#L1
    sources/github-repo.json#L1 private:false spdx_id MIT
    The runner repository is public with an MIT LICENSE and runs in the user's cluster; PARTIAL: 'No credit card' has no checkable artefact.
    C-215Governed MCP gateway. One connection point, full governance. Production [...] stored in your own database or hosted by OneDroid. Available with OneDroid Synapse [...] and evidence anyone can replay. Sign up to evaluate [...] Free and open — we help enterprises adopt it. Open source [...] on your infrastructure. Activeonedroid.ai/llms-full.txt (sources/llms-full.txt L4076-4116 (PAGE /synapse; under "OneDroid Synapse"))https://synapse.onedroid.ai/#200
    https://api.github.com/orgs/OneDro1d/repos#200 dark-factory Apache-2.0, argus-runner MIT
    sources/llms-full.txt#L4076-4116
    The gateway answers live, the Argus runner is public MIT and a public dark-factory kit repo is Apache-2.0; PARTIAL: the 'Production' and 'Active' status labels and Engram's storage options have no artefact here.
    C-218Claude Desktop and claude.ai connect through a consent screen instead of a token. In the app, Settings → Connectors, add a custom connector with `https://synapse.onedroid.ai/hub/<your-slug>/mcp`.onedroid.ai/llms-full.txt (sources/llms-full.txt L4430-4432 (PAGE /quickstart; under "Not in a terminal?"))https://synapse.onedroid.ai/hub/example/mcp#401 WWW-Authenticate resource_metadata .well-known/oauth-protected-resource/hub/example/mcp
    https://synapse.onedroid.ai/.well-known/oauth-authorization-server#200 authorization_endpoint /oauth/authorize, registration_endpoint /oauth/register
    The hub MCP path answers 401 with an OAuth protected-resource pointer and the server publishes OAuth metadata with authorize and dynamic-registration endpoints, the consent flow a custom connector uses.
    C-222Browser checks run as functional end-to-end scenarios.onedroid.ai/llms-full.txt (sources/llms-full.txt L5283-5284 (PAGE /qa-teams; under "Not load on a web page"))repo/testkit/ui/playwright.config.ts
    repo/internal/argus/ui_scenario.go#L38
    repo/README.md#L146
    The open runner ships a Playwright harness that ui scenarios drive, so browser checks run as end-to-end scenarios.
    C-225The applications set strictly necessary cookies for sign-in and session security (via Clerk) — nothing else. Walkthrough videos on the quickstart page are embedded from YouTube in its privacy-enhanced mode (youtube-nocookie.com): nothing is set until you press play [...] Fonts are loaded from Google Fonts, which receives standard request data when the page loads.onedroid.ai/llms-full.txt (sources/llms-full.txt L6458-6467 (PAGE /privacy; under "What we collect"))
    onedroid.ai/privacy.md (sources/privacy.md L67-76 (under "What we collect"))
    https://onedroid.ai/quickstart#200 L174 L250 L331 youtube-nocookie.com/embed
    https://onedroid.ai/quickstart#200 L18-20 fonts.googleapis.com
    The live quickstart embeds its three videos from youtube-nocookie.com and loads Google Fonts as stated; PARTIAL: 'only Clerk strictly necessary cookies' and 'nothing set until play' were not observed (no signed-in session, no browser).
    C-232Explicit allowances for AI crawlers alongside conventional search engines, and a sitemap covering every public page.onedroid.ai/llms-full.txt (sources/llms-full.txt L6308-6309 (PAGE /changelog; under "Crawlable: robots.txt and sitemap.xml"))
    onedroid.ai/changelog.md (sources/changelog.md L199-200 (under "Crawlable: robots.txt and sitemap.xml"))
    https://onedroid.ai/robots.txt#200 L1-28 Allow for GPTBot OAI-SearchBot ChatGPT-User ClaudeBot Claude-User PerplexityBot Google-Extended CCBot beside Googlebot
    https://onedroid.ai/sitemap.xml#200 30 loc
    robots.txt names AI crawlers alongside Googlebot and points to a sitemap that answers 200; PARTIAL: whether the 30-URL sitemap covers every public page was not checked exhaustively.
    C-233The governed gateway has moved to its own page at /synapse; nothing else was removed.onedroid.ai/llms-full.txt (sources/llms-full.txt L6270-6271 (PAGE /changelog; under "The site now leads with OneDroid Argus"))
    onedroid.ai/changelog.md (sources/changelog.md L161-162 (under "The site now leads with OneDroid Argus"))
    https://onedroid.ai/synapse#200 text/html
    https://onedroid.ai/sitemap.xml#200 L22 onedroid.ai/synapse
    The /synapse page answers 200 and is in the sitemap; PARTIAL: 'nothing else was removed' has no artefact.
    C-235Evidence anyone can replay. The withheld scenario set is hashed and written to a cryptographic ledger before the build starts.onedroid.ai/llms-full.txt (sources/llms-full.txt L114-115 (PAGE /; under "Same modernization, three setups: what the tester checks decides whether it helps"))repo/README.md#L117-132
    https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#200 block 47817413
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#200 block 47817439
    A public chain record shows a hash-mode ledger write at the seal before the verdict write, and the open CLI replays revealed scenarios; PARTIAL: 'before the build starts' is not shown, only before the run's verdict.
    C-236Every checkable claim this site makes, and what backs it: onedroid.ai/evidence/onedroid.ai.onedroid.ai/llms-full.txt (sources/llms-full.txt L46 (PAGE /; under "Test the agents where they work: in the workflow, in the fleet, next to the people who sign off."))https://onedroid.ai/evidence/onedroid.ai#200 sha256 48977c8fe050c9f8…
    The evidence page answers 200; PARTIAL: 'every checkable claim' cannot be checked for completeness.
    C-237Holdout acceptance. Scenarios sealed before the run, against a pinned build.onedroid.ai/llms-full.txt (sources/llms-full.txt L108 (PAGE /; under "Same modernization, three setups: what the tester checks decides whether it helps"))repo/docs/CERTIFICATE.md#L10-16
    repo/docs/CERTIFICATE.md#L85
    https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#200 block 47817413
    The certificate format binds a final run to a sealed set and a declared artifact digest, and a public seal tx precedes the verdict tx.
    C-238Under load. The same scenarios under load, against a message broker or an HTTP target; on an HTTP target a stepped ramp finds its limit.onedroid.ai/llms-full.txt (sources/llms-full.txt L110 (PAGE /; under "Same modernization, three setups: what the tester checks decides whether it helps"))repo/internal/amqpload/amqpload.go
    repo/internal/httpload/httpload.go#L174-218
    repo/templates/amqp-load.jmx
    The open runner has a broker load path (amqpload) and an HTTP stepped ramp (httpload) whose verdict reports the step where it stopped.
    C-240every scenario is revealed — trigger, expectation, evidence hash, verdict — and the runner replays any of them against the same build.onedroid.ai/llms-full.txt (sources/llms-full.txt L115-117 (PAGE /; under "Same modernization, three setups: what the tester checks decides whether it helps"))repo/README.md#L117-132
    repo/internal/reveal/replay.go
    repo/internal/reveal/verify.go
    The open repo documents the reveal document (every sealed scenario, Merkle proofs, anchors) and implements argus anchor verify and argus replay against a digest.
    C-241The control plane seals the set and anchors its hash before the run.onedroid.ai/llms-full.txt (sources/llms-full.txt L178-179 (PAGE /; under "Your agent writes the scenarios"))repo/docs/CERTIFICATE.md#L10-16
    https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#200 block 47817413 2026-10-07T19:58:34Z
    The format requires a sealed set anchor (version 2) and a public seal tx precedes the verdict tx on Base Sepolia; the sealing service itself is closed.
    C-242The coding agent never holds a token that can read the expectations, and the scenarios are never in its working directory.onedroid.ai/llms-full.txt (sources/llms-full.txt L185-186 (PAGE /; under "The builder never sees the answer key"))repo/internal/role/role.go#L1-20
    repo/internal/role/role.go#L52-56
    repo/internal/onboard/guard.go#L14-51
    The product hat's token has no scenario access and its bundles redact expected values, and the onboarder refuses a product folder that holds any scenario file.
    C-243Every verdict comes with its boundary. The reveal lists every check that was run. The certificate says what it proves and what it does not, for instance whether the running build was measured. Our case studies also list what was not tested.onedroid.ai/llms-full.txt (sources/llms-full.txt L132-134 (PAGE /; under "50,000 lines overnight. How do you know what it did?"))repo/README.md#L119-120
    repo/internal/certificate/verify.go#L300
    sources/case-study-finos-calm.md#L143
    The reveal holds every sealed scenario, the verifier prints a NOT MEASURED line when the running build was not measured, and the case studies carry a 'What this test did not cover' section.
    C-244Your own agent, with our skill, turns that into a scenario set: the happy path, the repeat, the wrong actor, the wrong state, every external call's failure, the race at the shared resource, and the load variant.onedroid.ai/llms-full.txt (sources/llms-full.txt L176-178 (PAGE /; under "Your agent writes the scenarios"))repo/skills/scenario-author/SKILL.md#L1-8
    The open scenario-author skill turns plain intent into valid scenarios; PARTIAL: the seven listed scenario kinds (repeat, wrong actor, race and so on) are not enumerated in the public skill.
    C-246A check that could not run never counts as a pass; a load run can be degraded; a comparison run reports differences, not a pass.onedroid.ai/llms-full.txt (sources/llms-full.txt L165 (PAGE /; under "You already had the incident."))repo/internal/argus/tally.go#L20-39
    repo/internal/compare/result.go#L11
    The tally counts error, not_applicable and degraded apart from passed, and compare's verdict vocabulary is differences, not a pass.
    C-250The same checks run under load against a message broker or an HTTP target, and on a schedule, every 15 minutes at the shortest.onedroid.ai/llms-full.txt (sources/llms-full.txt L2791 (PAGE /argus; under "Load and continuous runs come with it."))
    onedroid.ai/argus.md (sources/argus.md L55 (under "Load and continuous runs come with it."))
    repo/internal/amqpload/amqpload.go
    repo/internal/httpload/httpload.go#L174-218
    repo/skills/scenario-author/SKILL.md#L267-289
    Load against a broker and an HTTP target is implemented in the open runner; PARTIAL: the schedule and its 15-minute floor are in the closed control plane.
    C-253the set, and its hash goes on a cryptographic ledger before the work is judged, so anyone can check that the scenarios existed first and never changedonedroid.ai/llms-full.txt (sources/llms-full.txt L2818-2820 (PAGE /argus; under "Sealed before the build."))
    onedroid.ai/argus.md (sources/argus.md L82-84 (under "Sealed before the build."))
    https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#200 block 47817413 2026-10-07T19:58:34Z
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#200 block 47817439 2026-10-07T19:59:26Z
    repo/README.md#L117-130
    A public seal record precedes the verdict record on Base Sepolia, and argus anchor verify recomputes the set's Merkle root against the anchor so anyone can check it did not change.
    C-254The ledger proves the order of events and that nothing was edited. Running it again is what proves the result is true.onedroid.ai/llms-full.txt (sources/llms-full.txt L2855-2856 (PAGE /argus; under "The reveal"))
    onedroid.ai/argus.md (sources/argus.md L119-120 (under "The reveal"))
    repo/README.md#L117-132
    repo/internal/reveal/replay.go
    anchor verify checks order and integrity against the chain and replay re-runs a revealed scenario against a digest, the two separate checks the claim describes.
    C-255A ramp that crosses the limit at its top step has measured the limit, which is not a failure.onedroid.ai/llms-full.txt (sources/llms-full.txt L2871-2873 (PAGE /argus; under "Under load."))
    onedroid.ai/argus.md (sources/argus.md L135-137 (under "Under load."))
    repo/internal/httpload/httpload.go#L174-218
    The HTTP ramp verdict rule says a ramp that breaks above its first step is a measured limit, not a failure.
    C-256Under load, expectations are tolerances, and the report says so.onedroid.ai/llms-full.txt (sources/llms-full.txt L2875-2876 (PAGE /argus; under "Under load."))
    onedroid.ai/argus.md (sources/argus.md L139-140 (under "Under load."))
    repo/skills/scenario-author/SKILL.md#L267-289
    repo/skills/scenario-author/SKILL.md#L419
    A LOAD profile judges a Target P95 Ms and a Max Error Rate (and a Not Worse Than band) rather than exact values; PARTIAL: the report's wording that it judged tolerances was not traced.
    C-257A target takes load only after its operator lists it as one that may, and a live system should never be listed.onedroid.ai/llms-full.txt (sources/llms-full.txt L2879-2880 (PAGE /argus; under "Under load."))
    onedroid.ai/argus.md (sources/argus.md L143-144 (under "Under load."))
    repo/internal/config/config.go#L345-349
    repo/internal/config/config.go#L1078-1080
    Broker load is allowed only on targets the operator lists in load_allowed_targets, which validation checks; PARTIAL: 'a live system should never be listed' is advice, not enforced.
    C-258The same sealed checks can run against two systems, or two versions of one, and their recorded outputs are compared.onedroid.ai/llms-full.txt (sources/llms-full.txt L2898-2899 (PAGE /argus; under "One sealed set, two systems."))
    onedroid.ai/argus.md (sources/argus.md L162-163 (under "One sealed set, two systems."))
    repo/internal/compare/result.go#L11
    repo/internal/compare/output_read.go#L62
    repo/skills/scenario-author/SKILL.md#L406-419
    The open runner's compare package reads recorded outputs of two systems and gives an identical/differences verdict.
    C-260One skill sets OneDroid Argus up next to your system. Another turns a plain description of what should happen into valid scenarios.onedroid.ai/llms-full.txt (sources/llms-full.txt L2927-2928 (PAGE /argus; under "Agent skills"))
    onedroid.ai/argus.md (sources/argus.md L191-192 (under "Agent skills"))
    repo/skills/scenario-author/SKILL.md#L1-8
    repo/onboarding/HOW-TO-ARGUS-CONFIG.md
    The scenario-author skill (description to valid scenarios) is in the open repo; PARTIAL: a separate setup skill was not found there, only an onboarding how-to.
    C-261Released — open source, MITonedroid.ai/llms-full.txt (sources/llms-full.txt L2923 (PAGE /argus; under "CLI"))
    onedroid.ai/argus.md (sources/argus.md L187 (under "CLI"))
    repo/LICENSE#L1
    sources/github-repo.json#L1 visibility public, license MIT
    The CLI's repository is public with an MIT LICENSE, as GitHub's own metadata also reports.
    C-263Scenarios stay hidden from the builder until the tester reveals them, so nothing can be optimised for them.onedroid.ai/llms-full.txt (sources/llms-full.txt L2963-2964 (PAGE /argus; under "Not a test generator you keep"))
    onedroid.ai/argus.md (sources/argus.md L227-228 (under "Not a test generator you keep"))
    repo/internal/role/role.go#L17-20
    repo/internal/role/role.go#L52-56
    repo/internal/onboard/guard.go#L14-51
    The builder's hat has no scenario access and its outputs omit scenarios and expected values; PARTIAL: 'until the tester reveals them' rests on the closed reveal service.
    C-264A monitor can watch a live system, on its owner's terms.onedroid.ai/llms-full.txt (sources/llms-full.txt L2968 (PAGE /argus; under "Not a free hand on a live system"))
    onedroid.ai/argus.md (sources/argus.md L232 (under "Not a free hand on a live system"))
    repo/internal/config/config.go#L331
    repo/internal/config/moneywrites.go#L40-64
    The owner's config declares money_handling and the money_writes.allow list that bound what checks may write; PARTIAL: the monitor schedule itself is in the closed control plane.
    C-274A load check says the candidate must not be worse than the reference by more than a declared band, here 25%.onedroid.ai/llms-full.txt (sources/llms-full.txt L3256 (PAGE /case-study-otrs-znuny; under "What was checked"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L79 (under "What was checked"))
    repo/skills/scenario-author/SKILL.md#L419
    repo/internal/compare/judge.go#L14-L21
    sources/case-study-otrs-znuny.md#L79
    The `Not Worse Than` band against the reference median is defined in the authoring skill and as BandState not_worse/worse in the compare code; PARTIAL: the 25% band used in this study is not published in any check file.
    C-287Executor v0.3.66 on the same cluster, with read-only access to the demo's pods so that a certificate can say what was running.onedroid.ai/llms-full.txt (sources/llms-full.txt L3434 (PAGE /case-study-finos-calm; under "The setup"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L61 (under "The setup"))
    https://base-sepolia.blockscout.com/api/v2/blocks/47817384/transactions#fetched/007.txt raw_input decodes to {"image_digests":["sha256:21bc0b16…","sha256:2ec2a457…","sha256:feb820a3…"],"runner_digest":"sha256:5e5542b7…"}
    repo/internal/k8srender/k8srender.go#L2655-2669
    repo/onboarding/HOW-TO-ARGUS-CONFIG.md#L325
    The runner renders a get/list-only pod Role so a certifying run can measure running image digests, and the run's pre-run commitment on Base Sepolia records the three image digests and a runner digest; PARTIAL: the v0.3.66 version is not tied to that runner digest, and the cluster's grant itself is not shown.
    C-288The certified set holds seven checks, sealed on 7 October 2026 at 19:57 UTC.onedroid.ai/llms-full.txt (sources/llms-full.txt L3444 (PAGE /case-study-finos-calm; under "What was checked"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L71 (under "What was checked"))
    https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#fetched/005.txt block_number 47817413 timestamp 2026-10-07T19:58:34Z raw_input payload scenario_count:7
    https://base-sepolia.blockscout.com/api/v2/blocks/47817384/transactions#fetched/007.txt timestamp 2026-10-07T19:57:36Z commitment hc_ed5e3a334510d944f0
    sources/case-study-finos-calm.md#L71
    The seal anchor at block 47817413 carries set_hash and scenario_count 7 on 7 Oct 2026, chained to a commitment anchored at 19:57:36Z; PARTIAL: the set_hash transaction itself is timestamped 19:58:34Z, so the exact 19:57 seal minute rests on the commitment record.
    C-289No check was left not measured, and none could not run.onedroid.ai/llms-full.txt (sources/llms-full.txt L3480 (PAGE /case-study-finos-calm; under "The result"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L107 (under "The result"))
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#fetched/006.txt raw_input payload tallies {passed:5,failed:2,errored:0,degraded:0} verdict failed
    https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#fetched/005.txt scenario_count:7
    The anchored verdict's tallies (5+2+0+0) add up to the anchored scenario_count of 7 with errored 0, so every check has a pass or fail outcome on chain.
    C-292Before the run, the sealed set was recorded on three chains: a private chain run by OneDroid Argus, Base Sepolia (a public Ethereum test network) and Bitcoin, through OpenTimestamps. After the run, the verdict was recorded on the same three.onedroid.ai/llms-full.txt (sources/llms-full.txt L3500 (PAGE /case-study-finos-calm; under "How anyone can check it later"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L127 (under "How anyone can check it later"))
    https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#fetched/005.txt status ok block 47817413 set_hash
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#fetched/006.txt status ok block 47817439 verdict failed
    repo/README.md#L134
    Both quoted Base Sepolia transactions exist at the stated blocks, the seal before the verdict; PARTIAL: the private-chain and Bitcoin/OpenTimestamps records are not published (the README says the private chain reads UNREACHABLE from outside).
    C-293What goes on a chain is only ids, hashes, digests, counts and the verdict, never a check, an expected value or an observed one.onedroid.ai/llms-full.txt (sources/llms-full.txt L3500 (PAGE /case-study-finos-calm; under "How anyone can check it later"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L127 (under "How anyone can check it later"))
    https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#fetched/005.txt raw_input keys holdout_root, criteria_hash, set_hash, scenario_count
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#fetched/006.txt raw_input keys run_id, artifact_digest, evidence_bundle_hash, verdict, tallies, finished_at
    The decoded Base Sepolia inputs hold only ids, hashes, digests, counts, the verdict and a finish time, and no check body or expected or observed value; PARTIAL: the private-chain and Bitcoin records were not inspected.
    C-294A certificate built from these records can be checked with the `argus` command line, with no account and without seeing a single checkonedroid.ai/llms-full.txt (sources/llms-full.txt L3508 (PAGE /case-study-finos-calm; under "How anyone can check it later"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L135 (under "How anyone can check it later"))
    repo/README.md#L103-115
    repo/cmd/argus/certificate_cmd.go#L50
    repo/cmd/argus/certificate_cmd.go#L128
    `argus certificate verify` is registered in the open-source CLI and the README states it never contacts the control plane and reads anchors through the user's own chain client; PARTIAL: `certificate get` itself needs an author session, and this run's certificate is not published.
    C-336Claude Code, Cursor, the Agent SDKs — anything that speaks MCP connects without custom integration.onedroid.ai/llms-full.txt (sources/llms-full.txt L3973-3974 (PAGE /synapse; under "Connect"))https://synapse.onedroid.ai/hub/work/mcp#401 Www-Authenticate Bearer resource_metadata=.well-known/oauth-protected-resource/hub/work/mcp; Access-Control-Allow-Headers Mcp-Protocol-Version, Mcp-Session-Id
    sources/llms-full.txt#L3973-3978
    The quoted hub URL answers as a standard MCP streamable-HTTP endpoint with OAuth protected-resource discovery, so a generic MCP client can attach; PARTIAL: no connection by Cursor or the Agent SDKs is shown.
    C-349Sign in at synapse.onedroid.ai with Google or an email code.onedroid.ai/llms-full.txt (sources/llms-full.txt L4405-4406 (PAGE /quickstart; under "Create a hub"))https://clerk.onedroid.ai/v1/environment#fetched/008.txt first_factors [email_code, email_link, google_one_tap, oauth_google, password, …]; social.oauth_google.enabled true
    https://synapse.onedroid.ai#200
    OneDroid's sign-in environment enables Google OAuth and an email code; PARTIAL: the record names the application 'OneDroid', so its binding to synapse.onedroid.ai is inferred, not shown.
    C-350Watch it done: the Synapse walkthrough, recorded against the live service, at youtu.be/WJvsBV4sXFU.onedroid.ai/llms-full.txt (sources/llms-full.txt L4459-4460 (PAGE /quickstart; under "The check"))https://youtu.be/WJvsBV4sXFU#fetched/009.txt L1 'OneDroid Synapse walkthrough: one governed MCP gateway for all your agents'
    https://youtu.be/WJvsBV4sXFU#fetched/009.body ownerChannelName OneDroid_ai uploadDate 2026-09-25
    The video exists on the OneDroid_ai channel with a matching title; PARTIAL: 'recorded against the live service' cannot be checked from the page metadata.
    C-352Watch it done: the Engram walkthrough, recorded against the live service, at youtu.be/1o5mPIGCngk.onedroid.ai/llms-full.txt (sources/llms-full.txt L4524-4525 (PAGE /quickstart; under "Every write is a commit"))https://youtu.be/1o5mPIGCngk#fetched/010.txt L1 'OneDroid Engram walkthrough: versioned agent memory in a Postgres you own'
    https://youtu.be/1o5mPIGCngk#fetched/010.body ownerChannelName OneDroid_ai uploadDate 2026-09-25
    The video exists on the OneDroid_ai channel with a matching title; PARTIAL: 'recorded against the live service' cannot be checked from the page metadata.
    C-353The kits are the method OneDroid Argus enforces, written down as skills, hooks and gates any capable agent harness can run.onedroid.ai/llms-full.txt (sources/llms-full.txt L4536-4537 (PAGE /quickstart; under "Run your agent under a method you can audit."))https://api.github.com/repos/OneDro1d/dark-factory/git/trees/HEAD?recursive=1#200 paths skills/*/SKILL.md, hooks/df-stage-gate.py, plugins/df-governed/hooks/claim-gate.py, dispatch-gate.py, merge-gate.py
    The public repository holds the kits as skills, hooks and gate scripts; PARTIAL: that this is 'the method OneDroid Argus enforces' and runs on any harness is not shown.
    C-354The repository is github.com/OneDro1d/dark-factory (public, Apache-2.0).onedroid.ai/llms-full.txt (sources/llms-full.txt L4544-4546 (PAGE /quickstart; under "Clone and list the kits"))https://api.github.com/repos/OneDro1d/dark-factory#200 visibility public, license.spdx_id Apache-2.0
    https://github.com/OneDro1d/dark-factory#200
    GitHub's API reports the repository as public with an Apache-2.0 licence.
    C-355The bootstrap writes a new directory beside the checkout, pinned to the method at today's commit.onedroid.ai/llms-full.txt (sources/llms-full.txt L4559-4560 (PAGE /quickstart; under "Make your own copy"))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/bootstrap.sh#200 L16 'Defaults target-dir to ../<instance-name>, i.e. a sibling of this repo'
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/bootstrap.sh#L72 TARGET default ../$NAME, L98-102 resolve the Tier-1 pin via git ls-remote refs/heads/main
    The bootstrap script defaults its target to a sibling directory and pins the remote main commit SHA at run time into the lockfile.
    C-356For ChatGPT, Gemini or a plain API model, paste DARK-FACTORY-PRIMING.md into the conversation. It inlines the whole method — every stage skill and the adversary gate — so the agent needs nothing else.onedroid.ai/llms-full.txt (sources/llms-full.txt L4571-4574 (PAGE /quickstart; under "No skill system? One page"))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/docs/DARK-FACTORY-PRIMING.md#200 L1 'Self-Contained Priming Doc (for skill-less agents)'
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/docs/DARK-FACTORY-PRIMING.md#L147 '2.2 The adversary gate', L230 'Part 3 — The stages, fully inlined', L538 'Adaptation notes for skill-less agents (ChatGPT / Gemini / plain API)'
    The priming document exists, inlines the stages and the adversary gate, and addresses ChatGPT, Gemini and plain API models; PARTIAL: that it inlines every stage skill and needs nothing else was not checked line by line.
    C-357Stage skills for product owner, architect, test-first developer and QA; an adversary gate that checks evidence without redoing the work; dispatch patterns for parallel sub-agents; an autonomy policy for what stops for a human.onedroid.ai/llms-full.txt (sources/llms-full.txt L4584-4586 (PAGE /quickstart; under "What you get"))https://api.github.com/repos/OneDro1d/dark-factory/git/trees/HEAD?recursive=1#200 skills/df-product-owner/SKILL.md, skills/df-solution-architect/SKILL.md, skills/df-tdd-developer/SKILL.md, skills/df-qa/SKILL.md, skills/df-adversary-gate/SKILL.md, skills/df-dispatch-subagents/SKILL.md, skills/work-autonomously/SKILL.md
    Each named component is a SKILL.md in the public repository: product owner, architect, test-first developer, QA, the adversary gate, sub-agent dispatch and the autonomy policy.
    C-358Watch it done: the kits walkthrough, recorded on these pages with a synthesized narration, at youtu.be/KcbsT7jARFc.onedroid.ai/llms-full.txt (sources/llms-full.txt L4595-4596 (PAGE /quickstart; under "Where it leads"))https://youtu.be/KcbsT7jARFc#fetched/011.txt L1 'How do you run a coding agent under a method you can audit? The Dark Factory kits, in ten minutes'
    https://youtu.be/KcbsT7jARFc#fetched/011.body ownerChannelName OneDroid_ai uploadDate 2026-09-26
    The kits video exists on the OneDroid_ai channel; PARTIAL: 'recorded on these pages with a synthesized narration' cannot be checked from the metadata.
    C-359It does not decide what its tests may do: the tester and the system's owner do, the configuration records it, and OneDroid Argus enforces what the configuration declares.onedroid.ai/llms-full.txt (sources/llms-full.txt L4633-4635 (PAGE /security; under "Where the verifier runs, and where it never does."))
    onedroid.ai/security.md (sources/security.md L24-26 (under "Where the verifier runs, and where it never does."))
    repo/schemas/argus-config.schema.yaml#L78-85
    repo/internal/scenario/moneyguard.go#L15-L38
    repo/schemas/argus-config.schema.yaml#L196
    The open-source runner's config schema declares what checks may do (money_writes.allow, load_test: never), and the code refuses what the config does not list, at validate time and at run time; PARTIAL: who decides the configuration is a process statement.
    C-360A scheduled monitor can watch a live system.onedroid.ai/llms-full.txt (sources/llms-full.txt L4635-4636 (PAGE /security; under "Where the verifier runs, and where it never does."))
    onedroid.ai/security.md (sources/security.md L26-27 (under "Where the verifier runs, and where it never does."))
    repo/internal/report/report.go#L849
    repo/internal/k8srender/k8srender.go#L2658
    repo/docs/CERTIFICATE.md#L172
    The runner code has a `scheduled` certifying run mode next to `final` and `rehearsal`, and the certificate docs refer to later scheduled runs.
    C-361Its source is open under MIT at github.com/OneDro1d/argus-runner, so what it does can be read rather than taken on our word.onedroid.ai/llms-full.txt (sources/llms-full.txt L4640-4642 (PAGE /security; under "Where the verifier runs, and where it never does."))
    onedroid.ai/security.md (sources/security.md L31-33 (under "Where the verifier runs, and where it never does."))
    repo/LICENSE#L1-3
    sources/github-repo.json#license.spdx_id MIT, full_name OneDro1d/argus-runner, visibility public
    https://github.com/OneDro1d/argus-runner#200
    The public argus-runner repository carries an MIT LICENSE (Providentia Worldwide), and GitHub reports spdx_id MIT.
    C-366Tool-call metadata — which agent called which tool, on which connection, when, and whether it was allowed.onedroid.ai/llms-full.txt (sources/llms-full.txt L4676-4677 (PAGE /security; under "Hosted"))
    onedroid.ai/security.md (sources/security.md L67-68 (under "Hosted"))
    sources/security.md#L67-68
    https://synapse.onedroid.ai/assets/index-zeCwIjse.js#200 (routes /hubs/${s.id}/activity-logs, /hubs/${s.id}/audit-logs; log row fields timestamp, user_id, actor_type, tool, method, success, error_code, latency_ms)
    The hosted Synapse bundle reads activity-log rows carrying caller, tool, time and outcome fields; PARTIAL: an explicit allowed/denied field and the connection on each row were not seen in the bundle.
    C-368Connections are grouped into hubs with role-based membership, so access is granted to a team boundary rather than pasted into a config file.onedroid.ai/llms-full.txt (sources/llms-full.txt L4717-4718 (PAGE /security; under "Hubs and roles"))
    onedroid.ai/security.md (sources/security.md L108-109 (under "Hubs and roles"))
    sources/security.md#L108-109
    https://synapse.onedroid.ai/assets/index-zeCwIjse.js#200 (routes /hubs/${s.id}/members, /hubs/${s.id}/groups, /hubs/${s.id}/grants, /hubs/${s.id}/hub-connections; roles "admin", "member")
    The hosted Synapse app manages hub members with admin/member roles, groups, grants and hub connections, which implements hub-scoped role-based access to connections.
    C-370Our delivery method — including the gates, the enforcement hooks and the failures that produced them — is public under Apache-2.0.onedroid.ai/llms-full.txt (sources/llms-full.txt L4768-4770 (PAGE /security; under "How to verify us"))
    onedroid.ai/security.md (sources/security.md L159-161 (under "How to verify us"))
    sources/security.md#L159-161
    https://api.github.com/repos/OneDro1d/dark-factory#200 private=false visibility=public license.spdx_id=Apache-2.0
    https://api.github.com/repos/OneDro1d/dark-factory/git/trees/HEAD?recursive=1#200 (hooks/df-stage-gate.py, plugins/df-governed/hooks/claim-gate.py, boot-kit/scripts/publish-gate.sh)
    The dark-factory repository is public under Apache-2.0 and holds gate scripts and enforcement hooks; PARTIAL: 'the failures that produced them' has no identifiable file in the tree (only stages/3-Developer/03-why-we-test-first.md).
    C-371Engram is a memory layer in a Postgres you control with your own database, or one OneDroid hosts.onedroid.ai/llms-full.txt (sources/llms-full.txt L4804 (PAGE /engram; under "Portable, persistent context"))sources/llms-full.txt#L4804
    https://synapse.onedroid.ai/assets/index-zeCwIjse.js#200 ("Bring your own database", database_url, postgresql://, POST /hubs/test-db)
    https://engram.onedroid.ai/#200 text/html
    The hosted app offers a bring-your-own Postgres (database_url, a test-db call) next to the hosted Engram; PARTIAL: that Engram's memory layer actually lives in that database is not observable.
    C-374Scoped by group and namespace, so context is shared deliberately rather than pooled by accident — and, with your own database, in a Postgres you own; the library registry (which libraries exist and who belongs to them) stays in OneDroid's platform database.onedroid.ai/llms-full.txt (sources/llms-full.txt L4856-4857 (PAGE /engram; under "Permissioned and yours"))sources/llms-full.txt#L4856-4857
    https://engram.onedroid.ai/assets/api-DyQEmaGh.js#200 (routes /groups/${e}, /namespaces/${e}/grants, /libraries/${e}/grants, /namespaces/${e}/visibility)
    The Engram client scopes access by group, namespace and library grants; PARTIAL: content in a customer-owned Postgres and the registry staying in the platform database are not shown.
    C-375Every agent your team builds inherits the whole surface with a token — no new connector work, no new credential handling, no new review for each one.onedroid.ai/llms-full.txt (sources/llms-full.txt L4956-4958 (PAGE /financial-services; under "One integration, not one per agent"))
    onedroid.ai/financial-services.md (sources/financial-services.md L42-44 (under "One integration, not one per agent"))
    sources/financial-services.md#L42-44
    https://synapse.onedroid.ai/assets/index-zeCwIjse.js#200 (routes /hubs/${s.id}/tokens, /hubs/${s.id}/hub-connections)
    sources/llms.txt#L120-123
    Hub-scoped agent tokens over the hub's shared connections exist in the hosted app; PARTIAL: 'no new review for each one' is an organisational outcome not shown.
    C-387Acceptance criteria become a failing test list before code is written. Green means done; the tests never get weakened to pass.onedroid.ai/llms-full.txt (sources/llms-full.txt L5358-5359 (PAGE /dark-factory; under "Test-first discipline"))sources/llms-full.txt#L5358-5359
    https://api.github.com/repos/OneDro1d/dark-factory/git/trees/HEAD?recursive=1#200 (skills/df-tdd-developer/SKILL.md, stages/3-Developer/03-why-we-test-first.md, skills/test-quality-gatekeeper/SKILL.md)
    The public kits contain the test-first developer skill and a test-quality gatekeeper; PARTIAL: that tests are never weakened in real deliveries is not shown.
    C-390Decide which tools each hub, role, and agent can reach. Deny rules, tool grants, and data filters enforced at the gateway — not in a wiki.onedroid.ai/llms-full.txt (sources/llms-full.txt L5795-5796 (PAGE /enterprise; under "Policy enforcement"))sources/llms-full.txt#L5795-5796
    https://synapse.onedroid.ai/assets/index-zeCwIjse.js#200 (routes /hubs/${s.id}/grants, /hubs/${s.id}/tools/toggles, /hubs/${s.id}/groups)
    Per-hub tool grants, per-tool toggles and groups exist in the hosted app; PARTIAL: deny rules and data filters were not found in the bundle, and enforcement at the gateway is not observable.
    C-392A verdict in that record is never a score: each scenario is passed, failed, errored or not_applicable (it could not run; a load run has one more outcome, degraded), and neither an error nor not_applicable is ever a pass.onedroid.ai/llms-full.txt (sources/llms-full.txt L5859-5860 (PAGE /enterprise; under "One primitive: the evidence record"))repo/internal/report/report.go#L389-404 (StatusDegraded, StatusNotApplicable)
    repo/internal/report/report.go#L574 (StatusError)
    repo/internal/certificate/verify.go#L183-189
    The public code defines the per-scenario outcomes including degraded for load and not_applicable, and the verifier refuses a passed verdict that rests on not_applicable alone.
    C-393The OneDroid Argus runner is free.onedroid.ai/llms-full.txt (sources/llms-full.txt L5891 (PAGE /pricing))
    onedroid.ai/pricing.md (sources/pricing.md L4 (under "OneDroid Pricing — Free Runner, Free Gateway, Enterprise Licence"))
    repo/LICENSE#L1 (MIT License)
    sources/github-repo.json#L1 private=false visibility=public
    The runner's public repository carries an MIT licence.
    C-394The runner is free. The control plane is the product.onedroid.ai/llms-full.txt (sources/llms-full.txt L5906 (PAGE /pricing; under "The runner is free. The control plane is the product."))
    onedroid.ai/pricing.md (sources/pricing.md L21 (under "The runner is free. The control plane is the product."))
    repo/LICENSE#L1 (MIT License)
    sources/argus-runner-README.md#L9-10
    sources/pricing.md#L21
    The runner is MIT and public, and the README says the control plane is a hosted service not in the repository; PARTIAL: 'the control plane is the product' is a pricing statement.
    C-395Executes any revealed scenario, so anyone who runs it can replay a verdictonedroid.ai/llms-full.txt (sources/llms-full.txt L5916 (PAGE /pricing; under "The OneDroid Argus runner"))
    onedroid.ai/pricing.md (sources/pricing.md L31 (under "The OneDroid Argus runner"))
    repo/cmd/argus/reveal_cmd.go#L42 (flag set "replay")
    sources/argus-runner-README.md#L125-133
    The open CLI registers `argus replay <scenario id> --reveal reveal.json --against sha256:<digest>`, which runs a revealed scenario and compares it with the declared verdict.
    C-396Seals the scenario set your tester wrote and keeps it from the builderonedroid.ai/llms-full.txt (sources/llms-full.txt L5930 (PAGE /pricing; under "The OneDroid Argus control plane"))
    onedroid.ai/pricing.md (sources/pricing.md L45 (under "The OneDroid Argus control plane"))
    sources/case-study-finos-calm.md#L131
    https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#fetched/005.body block_number 47817413
    repo/internal/runner/limitedreport.go#L15-16
    A sealed set is anchored on Base Sepolia and the executor cuts the builder-facing report to a keep list; PARTIAL: that the builder's credential cannot read the set is a hosted control-plane rule not in the public code.
    C-397Seals it and anchors the hash before the build startsonedroid.ai/llms-full.txt (sources/llms-full.txt L5932 (PAGE /pricing; under "The OneDroid Argus control plane"))
    onedroid.ai/pricing.md (sources/pricing.md L47 (under "The OneDroid Argus control plane"))
    https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#fetched/005.body block_number 47817413 timestamp 2026-10-07T19:58:34Z
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#fetched/006.body block_number 47817439 timestamp 2026-10-07T19:59:26Z
    sources/case-study-finos-calm.md#L131-133
    The FINOS seal transaction is on chain before the verdict transaction; PARTIAL: the chain orders seal before verdict, not seal before the start of a build.
    C-398Adjudicates the final run against a pinned artefactonedroid.ai/llms-full.txt (sources/llms-full.txt L5934 (PAGE /pricing; under "The OneDroid Argus control plane"))
    onedroid.ai/pricing.md (sources/pricing.md L49 (under "The OneDroid Argus control plane"))
    repo/docs/CERTIFICATE.md#L95
    sources/argus-runner-README.md#L131-133
    repo/internal/certificate/verify.go#L292
    The certificate records the declared artefact digest and a run whose digest is not running is refused, and replay is bound to the certified digest; PARTIAL: the adjudication itself runs in the hosted control plane.
    C-400Your tester writes the scenarios with our open skillonedroid.ai/llms-full.txt (sources/llms-full.txt L269 (PAGE /; under "Open where it has to be."))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L5941 (PAGE /pricing; under "The OneDroid Argus control plane"))
    onedroid.ai/pricing.md (sources/pricing.md L56 (under "The OneDroid Argus control plane"))
    repo/skills/scenario-author/SKILL.md#L20
    repo/LICENSE#L1 (MIT License)
    sources/pricing.md#L56
    The scenario-author skill is in the public MIT repository.
    C-401The holdout is provably prior, and every scenario is replayable with the runner in your own environment.onedroid.ai/llms-full.txt (sources/llms-full.txt L5942-5943 (PAGE /pricing; under "The OneDroid Argus control plane"))
    onedroid.ai/pricing.md (sources/pricing.md L57-58 (under "The OneDroid Argus control plane"))
    https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#fetched/005.body block_number 47817413
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#fetched/006.body block_number 47817439
    repo/cmd/argus/reveal_cmd.go#L42
    The seal anchor precedes the verdict anchor on a public chain and the open CLI replays revealed scenarios; PARTIAL: being prior to the build, as against prior to the verdict, is not shown by the chain.
    C-402Bring your own database — your Engram content stays in your Postgresonedroid.ai/llms-full.txt (sources/llms-full.txt L5964 (PAGE /pricing; under "Hosted, any team size"))
    onedroid.ai/pricing.md (sources/pricing.md L79 (under "Hosted, any team size"))
    sources/pricing.md#L79
    https://synapse.onedroid.ai/assets/index-zeCwIjse.js#200 ("Bring your own database", database_url, postgresql://)
    The hosted app offers a bring-your-own Postgres option; PARTIAL: that Engram content stays in it is not observable.
    C-411OneDroid Argus is the newest piece and now the front of the houseonedroid.ai/llms-full.txt (sources/llms-full.txt L6042-6043 (PAGE /about; under "Infrastructure for agents"))sources/llms-full.txt#L16-17 (PAGE / Title "OneDroid Argus — Testing and Live Audits of AI Agents, Workflows and Fleets")
    https://onedroid.ai/llms-full.txt#200 sha256 2b063c3ac9cf2b36
    The live home page is titled and led by OneDroid Argus; PARTIAL: that Argus is the newest piece has no dated artefact.
    C-414summaries are never accepted — only evidence a human or another agent can independently re-checkonedroid.ai/llms-full.txt (sources/llms-full.txt L6078-6079 (PAGE /about; under "You can read the method before you trust it."))sources/llms-full.txt#L6078-6079
    https://api.github.com/repos/OneDro1d/dark-factory/git/trees/HEAD?recursive=1#200 (skills/df-adversary-gate/SKILL.md, plugins/df-governed/hooks/claim-gate.py)
    The public method contains an adversary-gate skill and a claim-gate hook that reject self-reports; PARTIAL: that summaries are never accepted in practice is not shown.
    C-4152026-10-20 ### OneDroid Argus 0.3.74: a check can listen on a message bus, and read the logs an environment already keepsonedroid.ai/llms-full.txt (sources/llms-full.txt L6126-6128 (PAGE /changelog; under "What actually shipped."))
    onedroid.ai/changelog.md (sources/changelog.md L17-19 (under "What actually shipped."))
    repo/internal/chain/bus.go#L15-18
    repo/internal/chain/logsstep.go#L163 (observability.loki.log_regex)
    repo/skills/scenario-author/SKILL.md#L20 (released executor v0.3.74)
    The bus chain step and the shared-Loki log reading are in the public code and the skill names v0.3.74 as released; PARTIAL: the 2026-10-20 date has no public release or tag.
    C-417The limits: the bus step reads JSON message bodies only and needs an executor at 0.3.74 or later.onedroid.ai/llms-full.txt (sources/llms-full.txt L6144 (PAGE /changelog; under "OneDroid Argus 0.3.74: a check can listen on a message bus, and read the logs an environment already keeps"))
    onedroid.ai/changelog.md (sources/changelog.md L35 (under "OneDroid Argus 0.3.74: a check can listen on a message bus, and read the logs an environment already keeps"))
    repo/internal/chain/bus.go#L15-18 (claims on ONE JSON document, undecodable counted)
    repo/skills/scenario-author/SKILL.md#L20
    The bus step judges claims on one JSON document and counts undecodable messages separately; PARTIAL: the 0.3.74 minimum rests only on the skill's version note, with no public release.
    C-4182026-10-19 ### OneDroid Argus 0.3.73: a report can leave out everything the system under test saidonedroid.ai/llms-full.txt (sources/llms-full.txt L6162-6164 (PAGE /changelog; under "The site now leads with what testing costs"))
    onedroid.ai/changelog.md (sources/changelog.md L53-55 (under "The site now leads with what testing costs"))
    repo/internal/runner/limitedreport.go#L15-16
    repo/internal/runner/limitedreport.go#L131-133 (limitedNotSent)
    The executor-side limited report keep list leaves out what the system under test returned; PARTIAL: the 2026-10-19 date and the 0.3.73 number have no public release artefact.
    C-419In every report, the executor now sends only OneDroid Argus's own words when a run or a relayed command fails, never the application's error text. Load-test errors reach the limited report and the results push only as a kind from a fixed list and a count; the load tool's own messages stay in the full report, which only a Full-access author can read.onedroid.ai/llms-full.txt (sources/llms-full.txt L6168-6171 (PAGE /changelog; under "OneDroid Argus 0.3.73: a report can leave out everything the system under test said"))
    onedroid.ai/changelog.md (sources/changelog.md L59-62 (under "OneDroid Argus 0.3.73: a report can leave out everything the system under test said"))
    repo/internal/runner/limitedreport.go#L15-16
    repo/internal/runner/limitedreport.go#L92-103 (load errors as kind and count)
    repo/internal/runner/limitedreport.go#L215-218 (never an underlying error's text)
    The public executor code sends load errors to the limited report only as a closed-list kind and a count and replaces error text with its own sentences.
    C-4202026-10-13 ### OneDroid Argus 0.3.72: a check can depend on what the environment has, and test an app behind Clerkonedroid.ai/llms-full.txt (sources/llms-full.txt L6181-6183 (PAGE /changelog; under "OneDroid Argus 0.3.73: a report can leave out everything the system under test said"))
    onedroid.ai/changelog.md (sources/changelog.md L72-74 (under "OneDroid Argus 0.3.73: a report can leave out everything the system under test said"))
    repo/internal/precondition/precondition.go#L1-7
    repo/internal/config/clerksession.go#L24 (ClerkSessionType "clerk_session")
    repo/docs/CERTIFICATE.md#L136 (v5: 0.3.72)
    Preconditions and the clerk_session auth type are in the public code and CERTIFICATE.md ties v5 to 0.3.72; PARTIAL: the 2026-10-13 date has no public release artefact.
    C-421A run where every check is `not_applicable` never reads as passed. The results are bound into a new certificate format, v5. Formats v1 to v4 are unchanged, and an older `argus certificate verify` refuses v5 by name.onedroid.ai/llms-full.txt (sources/llms-full.txt L6189-6191 (PAGE /changelog; under "OneDroid Argus 0.3.72: a check can depend on what the environment has, and test an app behind Clerk"))
    onedroid.ai/changelog.md (sources/changelog.md L80-82 (under "OneDroid Argus 0.3.72: a check can depend on what the environment has, and test an app behind Clerk"))
    repo/internal/certificate/verify.go#L171-172
    repo/internal/certificate/verify.go#L188-189
    repo/docs/CERTIFICATE.md#L80-81
    The verifier refuses a passed verdict where every scenario is not_applicable, v5 carries the anchored not_applicable tally, and the doc states older binaries refuse v5 by name.
    C-4222026-10-13 ### OneDroid Argus 0.3.71: the money guard can allow a clean-up, a UI drive or a fault, one by oneonedroid.ai/llms-full.txt (sources/llms-full.txt L6204-6206 (PAGE /changelog; under "OneDroid Argus 0.3.72: a check can depend on what the environment has, and test an app behind Clerk"))
    onedroid.ai/changelog.md (sources/changelog.md L95-97 (under "OneDroid Argus 0.3.72: a check can depend on what the environment has, and test an app behind Clerk"))
    repo/internal/scenario/moneyengines.go#L211 (MatchCleanup)
    repo/internal/scenario/moneyengines.go#L227 (MatchUI)
    repo/internal/scenario/moneyengines.go#L242 (MatchFault)
    The money-write allowlist has separate clean-up, UI and fault matchers in the public code; PARTIAL: the 2026-10-13 date and the 0.3.71 number have no public release artefact.
    C-4232026-10-11 ### OneDroid Argus 0.3.70: a check can read the database and the logs, and make the system failonedroid.ai/llms-full.txt (sources/llms-full.txt L6224-6226 (PAGE /changelog; under "OneDroid Argus 0.3.71: the money guard can allow a clean-up, a UI drive or a fault, one by one"))
    onedroid.ai/changelog.md (sources/changelog.md L115-117 (under "OneDroid Argus 0.3.71: the money guard can allow a clean-up, a UI drive or a fault, one by one"))
    repo/internal/chain/dbstep.go#file (db chain step)
    repo/internal/chain/logsstep.go#L163
    repo/internal/chain/fault.go#file (fault chain step)
    The db, logs and fault chain steps are in the public code and the skill ties the db step to 0.3.70; PARTIAL: the 2026-10-11 date has no public release artefact.
    C-424the repository has been public since 7 October, announced todayonedroid.ai/llms-full.txt (sources/llms-full.txt L6252 (PAGE /changelog; under "The OneDroid Argus runner is open source"))
    onedroid.ai/changelog.md (sources/changelog.md L143 (under "The OneDroid Argus runner is open source"))
    sources/github-repo.json#L1 created_at=2026-10-07T22:26:48Z private=false visibility=public
    sources/changelog.md#L143
    The GitHub API record shows the repository created on 7 October 2026 and public now; PARTIAL: creation date is not proof of the date it became public.
    C-4252026-09-10 ### The site now leads with OneDroid Argusonedroid.ai/llms-full.txt (sources/llms-full.txt L6264-6266 (PAGE /changelog; under "The OneDroid Argus runner is open source"))
    onedroid.ai/changelog.md (sources/changelog.md L155-157 (under "The OneDroid Argus runner is open source"))
    sources/llms-full.txt#L16-17 (PAGE / Title "OneDroid Argus — ...")
    sources/changelog.md#L155-157
    The live home page leads with OneDroid Argus; PARTIAL: the 2026-09-10 date has no artefact beyond the changelog entry.
    C-4272026-07-19 ### Privacy policy and terms of service publishedonedroid.ai/llms-full.txt (sources/llms-full.txt L6297-6299 (PAGE /changelog; under "Google OAuth application verification completed"))
    onedroid.ai/changelog.md (sources/changelog.md L188-190 (under "Google OAuth application verification completed"))
    https://onedroid.ai/privacy.md#200 (sources/privacy.md)
    https://onedroid.ai/terms#200 text/html ("Effective date: 19 July 2026")
    Both pages are published and the terms carry a 19 July 2026 effective date; PARTIAL: the publication date rests on the pages' own date lines.
    C-4282026-07-19 ### Crawlable: robots.txt and sitemap.xmlonedroid.ai/llms-full.txt (sources/llms-full.txt L6304-6306 (PAGE /changelog; under "Privacy policy and terms of service published"))
    onedroid.ai/changelog.md (sources/changelog.md L195-197 (under "Privacy policy and terms of service published"))
    https://onedroid.ai/robots.txt#200 text/plain
    https://onedroid.ai/sitemap.xml#200 application/xml (30 <loc> entries)
    robots.txt and sitemap.xml are served; PARTIAL: the 2026-07-19 date has no artefact.
    C-4292026-07-18 ### Bring-your-own-database for OneDroid Engramonedroid.ai/llms-full.txt (sources/llms-full.txt L6311-6313 (PAGE /changelog; under "Crawlable: robots.txt and sitemap.xml"))
    onedroid.ai/changelog.md (sources/changelog.md L202-204 (under "Crawlable: robots.txt and sitemap.xml"))
    sources/changelog.md#L202-207
    https://synapse.onedroid.ai/assets/index-zeCwIjse.js#200 ("Bring your own database", database_url, POST /hubs/test-db)
    A bring-your-own-database option exists in the hosted app; PARTIAL: the 2026-07-18 date and Engram content landing in it are not shown.
    C-4302026-07-18 ### onedroid.ai launchedonedroid.ai/llms-full.txt (sources/llms-full.txt L6319-6321 (PAGE /changelog; under "Bring-your-own-database for OneDroid Engram"))
    onedroid.ai/changelog.md (sources/changelog.md L210-212 (under "Bring-your-own-database for OneDroid Engram"))
    https://onedroid.ai#200 sha256 d1aa54a6db1d448c
    sources/changelog.md#L210-212
    onedroid.ai is live; PARTIAL: the 2026-07-18 launch date has no artefact beyond the changelog.
    C-443Three uses on one evidence contract: holdout acceptance, under load, and continuous.onedroid.ai/llms.txt (sources/llms.txt L81-82 (under "Product"))repo/internal/report/report.go#L389-395 (StatusDegraded, load mode)
    repo/docs/CERTIFICATE.md#L72 (passed + failed + errored + degraded + not_applicable)
    repo/skills/scenario-author/SKILL.md#L129 (author_set_schedule on a sealed set)
    Holdout and load runs share one report and certificate tally format in the public code; PARTIAL: continuous (scheduled) use is a hosted control-plane tool named in the skill, not shown in code.
    C-444Sign in to create a hub, connect accounts, and issue an agent token.onedroid.ai/llms.txt (sources/llms.txt L98 (under "Applications"))https://synapse.onedroid.ai/#200
    https://synapse.onedroid.ai/assets/index-zeCwIjse.js#200 (routes /hubs/${s.id}/hub-connections, /hubs/${s.id}/tokens)
    The Synapse app is live and its client creates hubs, connects accounts and issues hub tokens.
    C-445OneDroid's own claims register is in it once, as part of the home page.onedroid.ai/llms.txt (sources/llms.txt L129 (under "Optional"))sources/llms-full.txt#L320 (single 'Every checkable claim' register heading)
    sources/llms-full.txt#L16 (PAGE /) and #L2740 (next PAGE /argus)
    The saved llms-full.txt has exactly one register heading, inside the home-page block.
    C-446Checked 2026-10-28, by OneDroid on itself, with the same method as the evidence pages.onedroid.ai/llms.txt (sources/llms.txt L51 (under "Claims and evidence"))https://onedroid.ai/evidence/onedroid.ai.md#200 text/markdown (L10 "checked 2026-10-28 · on ourselves")
    The published self-evidence page is dated 2026-10-28 and marked as OneDroid on itself; PARTIAL: 'the same method as the evidence pages' is not independently shown.
    C-447Not part of llms-full.txt; each page has a Markdown twin at `<url>.md`.onedroid.ai/llms.txt (sources/llms.txt L131 (under "Optional"))sources/llms-full.txt#L16-6589 (page markers; none is '# PAGE: https://onedroid.ai/evidence')
    https://onedroid.ai/evidence/onedroid.ai.md#200 text/markdown
    https://onedroid.ai/evidence#200 text/html
    No evidence page is concatenated into llms-full.txt and the onedroid.ai evidence page answers at its .md twin; PARTIAL: the twin was checked for one page only, not each.
    C-448The builder gets a pass/fail report of what its system [...] for a failed check, a description of the observed behaviour and, for an HTTP check, the status it returned. It never gets the scenario's expected values, nor the log lines and response-body excerpts a tester sees, since those can carry what the check looks for.docs.onedroid.ai/argus (sources/argus-docs.html L103-106 (under "The holdout"))
    docs.onedroid.ai/argus.md (sources/argus-docs.md L56-59 (under "The holdout"))
    repo/internal/runner/limitedreport.go#L15-16
    repo/internal/runner/limitedreport.go#L80-91 (steps observed_status, failed_claims[].claim)
    sources/argus-docs.md#L56-59
    The executor-built limited report sends HTTP status codes and leaves out log lines and body excerpts; PARTIAL: it sends the failed claims as written, so 'never gets the expected values' is not backed.
    C-449A **control plane** holds the scenario catalog, the run history, and the workspace that scopes who can see what. An **execution plane** runs next to your system — inside its own network or namespace — and is the only thing that ever calls it.docs.onedroid.ai/argus.md (sources/argus-docs.md L83-85 (under "Two planes"))sources/argus-runner-README.md#L9-10
    sources/argus-runner-README.md#L44 (the runner connects out only)
    repo/internal/runner/client.go#L966-973
    The public runner is the execution plane that reports to a hosted control plane holding run state; PARTIAL: the control plane's catalog and workspaces are hosted, and 'the only thing that ever calls it' is not checkable.
    C-450The execution plane reaches the control plane outbound only: your system's network needs egress, never an inbound route.docs.onedroid.ai/argus (sources/argus-docs.html L127-128 (under "Two planes"))
    docs.onedroid.ai/argus.md (sources/argus-docs.md L85-86 (under "Two planes"))
    repo/internal/runner/client.go#L28-36
    repo/README.md#L44
    sources/argus-docs.md#L85-86
    The executor's federation client is written as an outbound-only HTTPS client to the control plane's /fed/* endpoints (client.go L28 'direct outbound-only HTTPS', L36 'the outbound-only mirror'), and the README says the runner needs no inbound route.
    C-453A run with any `not_applicable` check, or with precondition results, gets a version 5 certificate.docs.onedroid.ai/argus.md (sources/argus-docs.md L119-120 (under "What changed recently"))repo/internal/certificate/verify.go#L171
    repo/internal/certificate/certificate.go#L36
    repo/internal/certificate/certificate.go#L53
    FormatV5 'argus-certificate/v5' is defined for the precondition tier (certificate.go L36, L53) and verify.go L171 rejects any non-v5 certificate with a not_applicable count; PARTIAL: the version tag v0.3.x is not checkable publicly.
    C-454**Fault steps: break the system under test during a run** (v0.3.69): a `fault` step in a chain scenario can stop, start or scale a Deployment, wait for it, or delay, cut or truncate a connection through a Toxiproxy.docs.onedroid.ai/argus.md (sources/argus-docs.md L166-168 (under "What changed recently"))repo/internal/fault/clients.go#L16-29
    repo/internal/fault/manager.go#L39
    repo/internal/fault/manager.go#L330-339
    internal/fault sets a Deployment's replica count and waits for readiness (clients.go L16-22), and drives Toxiproxy toxics for latency, limit_data and an OpTruncate op (manager.go L39, L330-339), documented in docs/fault-steps.md; PARTIAL: the v0.3.69 release tag is not publicly checkable.
    C-455**HTTP load: a stepped ramp that measures a limit** (v0.3.66): a new `HTTP Load` layer raises users step by step against one http target you allowed, stops at the first step that is not comfortable, and the Capacity page shows the limit in users.docs.onedroid.ai/argus.md (sources/argus-docs.md L180-182 (under "What changed recently"))repo/internal/httpload/httpload.go#L1-2
    repo/internal/httpload/httpload.go#L177-215
    repo/internal/config/loadallowed.go#L10
    internal/httpload judges an HTTP Load ramp that stops at the first step that is not comfortable (httpload.go L177-215) against allowed targets; PARTIAL: the Capacity page in the hosted web app and the v0.3.66 tag are not in the public artefacts.
    C-456a chain scenario's `http` steps now send the run's `X-Correlation-Id`docs.onedroid.ai/argus.md (sources/argus-docs.md L185 (under "What changed recently"))repo/internal/argus/chain_scenario.go#L446-450
    chain_scenario.go sets headers['X-Correlation-Id'] = corr on chain http steps unless the scenario already set it (L446-450).
    C-457an MCP check against a server that sends notifications before its result (on a `text/event-stream` answer) now judges the result, not the first notification.docs.onedroid.ai/argus.md (sources/argus-docs.md L195-196 (under "What changed recently"))repo/internal/mcp/client.go#L367-370
    repo/internal/mcp/client.go#L406-415
    The MCP client branches on a text/event-stream answer and skips notifications (no id) sent before the JSON-RPC result, failing only if the stream ends without one (client.go L406-415).
    C-458**Clearer onboarding** (v0.3.63 and v0.3.65): credential files are created 0600, a full Docker address pool is named, and a workspace-bound token says it cannot onboard.docs.onedroid.ai/argus.md (sources/argus-docs.md L199-200 (under "What changed recently"))repo/cmd/argus/main.go#L221-224
    repo/cmd/argus/main.go#L1974
    repo/internal/onboard/cloud.go#L1074
    Credential and session files are written 0o600 (main.go L221-224, L1974) and onboarding refuses a workspace-bound token (cloud.go L1074); PARTIAL: no named full Docker address pool message was found in the public repo, and the v0.3.63/v0.3.65 tags are not checkable.
    C-459**Load checks keep connections open, and the load record says why requests failed** (v0.3.63)docs.onedroid.ai/argus.md (sources/argus-docs.md L202 (under "What changed recently"))repo/internal/argus/httpload_run.go#L136
    repo/internal/httpload/httpload.go#L104
    The HTTP load run sets http.keepalive=true ('a user is a persistent client', httpload_run.go L136) and the load record carries error_reasons, 'why the transport failures failed' (httpload.go L104); PARTIAL: the v0.3.63 tag is not checkable.
    C-460**A target can be declared never load tested** (v0.3.60): `load_test: never`.docs.onedroid.ai/argus.md (sources/argus-docs.md L214 (under "What changed recently"))repo/internal/config/neverload.go#L12-21
    repo/schemas/argus-config.schema.yaml#L174
    repo/onboarding/HOW-TO-ARGUS-CONFIG.md#L194
    `load_test: never` on a test_targets entry is parsed and enforced in internal/config/neverload.go and declared in the config schema L174; PARTIAL: the v0.3.60 tag is not checkable.
    C-461**Logs and metrics survive a pod recreate on Kubernetes** (v0.3.62): the Loki and Pushgateway now keep their data on volumes, and `--obs-storage-class` picks the class.docs.onedroid.ai/argus.md (sources/argus-docs.md L207-208 (under "What changed recently"))repo/internal/k8srender/k8srender.go#L1635
    repo/internal/k8srender/k8srender.go#L1717-1729
    repo/cmd/argus/main.go#L3410
    render-k8s emits observability PersistentVolumeClaims with an optional storageClassName (k8srender.go L1717-1729) and the --obs-storage-class flag is registered (main.go L3410); PARTIAL: the v0.3.62 tag is not checkable.
    C-462**`check_env`, `${INGESTION_URL}` and clearer onboarding errors** (v0.3.62)docs.onedroid.ai/argus.md (sources/argus-docs.md L211 (under "What changed recently"))repo/internal/config/checkenv.go#L15
    repo/internal/config/checkenv.go#L83
    repo/internal/argus/chain_scenario.go#L51
    check_env is parsed and limited in internal/config/checkenv.go and ${INGESTION_URL} is substituted in chain scenarios (chain_scenario.go L51, example scenario L9); PARTIAL: 'clearer onboarding errors' and the v0.3.62 tag are not separately evidenced.
    C-463**A blocked broker reads as what it is** (v0.3.55 and v0.3.56): the run fails with `blocked by broker: <reason>`.docs.onedroid.ai/argus.md (sources/argus-docs.md L221-222 (under "What changed recently"))repo/internal/amqpload/amqpload.go#L177
    repo/internal/amqpload/amqpload.go#L268
    repo/internal/amqpload/amqpload.go#L418
    amqpload parses '503 blocked by broker: <reason>' and fails the run with a 'blocked by broker' outcome (amqpload.go L177, L268, L418); PARTIAL: the v0.3.55/v0.3.56 tags are not checkable.
    C-464**A failed numeric claim shows the number** (v0.3.54): `failed_claims[].observed` holds the value the field held, not the saved-value placeholder.docs.onedroid.ai/argus.md (sources/argus-docs.md L223-224 (under "What changed recently"))repo/internal/runner/limitedreport.go#L137
    repo/internal/report/report.go#L90
    The report schema carries failed_claims with an observed field (limitedreport.go L137, report.go L90); PARTIAL: the 'not the saved-value placeholder' behaviour and the v0.3.54 tag are not shown beyond the field.
    C-465**Comparing systems** (v0.3.57): see [Comparing systems](/argus-compare).docs.onedroid.ai/argus.md (sources/argus-docs.md L225 (under "What changed recently"))https://docs.onedroid.ai/argus-compare#200 text/html
    repo/internal/compare/hash.go#L19
    The linked Comparing systems page answers 200 and the runner has an internal/compare package; PARTIAL: the v0.3.57 tag is not checkable.
    C-466**A least-privilege AMQP load login, and metrics retention** (v0.3.53): `queues.load` sets the queue name prefix, and `observability.pushgateway.group_retention` limits how long a run's metrics stay.docs.onedroid.ai/argus.md (sources/argus-docs.md L231-233 (under "What changed recently"))repo/internal/argus/amqpload_run.go#L61
    repo/internal/argus/amqpload_run.go#L175
    repo/internal/config/pushgateway.go#L53
    queues.load sets the AMQP load queue prefix (amqpload_run.go L61, L175) and observability.pushgateway.group_retention is parsed (pushgateway.go L53), with the least-privilege login described in HOW-TO L160; PARTIAL: the v0.3.53 tag is not checkable.
    C-469**AMQP load testing is open** (v0.3.52): a load run against a broker you list under `load_allowed_targets`.docs.onedroid.ai/argus.md (sources/argus-docs.md L239-240 (under "What changed recently"))repo/internal/config/loadallowed.go#L13
    repo/internal/argus/argus.go#L1082
    repo/internal/amqpload/amqpload.go#L6
    load_allowed_targets is parsed in internal/config/loadallowed.go and gates load runs (argus.go L1082), with AMQP load in internal/amqpload; PARTIAL: the v0.3.52 tag is not checkable.
    C-470**Log queries take a whole correlation id** (v0.3.52), for every role.docs.onedroid.ai/argus.md (sources/argus-docs.md L244 (under "What changed recently"))repo/internal/obsquery/corrid.go#L13-25
    corrid.go requires the whole correlation id to match ('no unanchored substring, no suffix: a prefix is the widening bug', L25); PARTIAL: 'for every role' and the v0.3.52 tag are not separately evidenced.
    C-471**`argus preflight` checks Docker and host ports** on a Docker machine, and `--obs none` renders no observability (v0.3.52)docs.onedroid.ai/argus.md (sources/argus-docs.md L249-250 (under "What changed recently"))repo/internal/preflight/preflight.go#L363-381
    repo/internal/preflight/preflight.go#L393
    repo/cmd/argus/main.go#L2953
    preflight checks that the Docker daemon answers (docker info, L363-381) and that the published host ports are free (L393), and --obs none is handled (main.go L2953); PARTIAL: the v0.3.52 tag is not checkable.
    C-472The CLI is the same binary as the in-environment test server.docs.onedroid.ai/argus-quickstart (sources/argus-quickstart.html L68 (under "1. Get the argus CLI"))
    docs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L17 (under "1. Get the `argus` CLI"))
    repo/cmd/argus/main.go#L1-2
    repo/cmd/argus/main.go#L186
    cmd/argus is one binary whose CLI and MCP server are two entry points over the same core (main.go L1-2), with `serve --mode runner` as the in-environment server (L186).
    C-473`version` reports this binary's own build identity (its version, commit and build date) and reads nothing else — no config, no scenarios, no control planedocs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L61-62 (under "1. Get the `argus` CLI"))repo/cmd/argus/main.go#L745-754
    The `version` command is handled above the token gate and emits only buildinfo values (version, commit, build date, built-in hosts), reading no config, scenarios or control plane (main.go L745-754).
    C-474That build has every command this page uses (`cloud-login`, `doctor`, `preflight`, `cloud-create-workspace`, `render-k8s`, `certificate`); it refuses only `serve --mode control`, which is the hosted part.docs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L22-24 (under "1. Get the `argus` CLI"))repo/cmd/argus/usage_commands.go#L8-41
    repo/cmd/argus/main.go#L2906
    The public build registers cloud-login, doctor, preflight, cloud-create-workspace, render-k8s and certificate (usage_commands.go) and refuses `serve --mode control` as 'not part of this build' (main.go L2906).
    C-475The open runner's source also compiles for Windows, but that build is not released and not tested, and the onboarding kit is a set of bash scripts.docs.onedroid.ai/argus-quickstart (sources/argus-quickstart.html L92-93 (under "1. Get the argus CLI"))
    docs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L48-49 (under "1. Get the `argus` CLI"))
    repo/cmd/argus/up_detach_windows.go
    repo/internal/runner/summarymetrics_errno_windows.go
    https://api.github.com/repos/OneDro1d/argus-runner/releases#200 (empty list [])
    Windows-specific source files exist in the public tree, the public repo publishes no release, and the README names the onboarding kit as onboarding/onboard.sh and compose/k3d scripts; PARTIAL: 'not tested' and the private release set cannot be checked.
    C-476From v0.3.41, `argus <command> --help` prints that command's own usage and exits `0`, with no token set.docs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L66-67 (under "1. Get the `argus` CLI"))repo/cmd/argus/help_filter.go#L11-34
    repo/cmd/argus/main.go#L484-486
    help_filter.go implements per-command `argus <cmd> --help` output and documents its place relative to the token gate (L11-34); PARTIAL: exit code 0 with no token and the v0.3.41 tag are not traced end to end.
    C-477Presenting `ARGUS_EXECUTOR_SECRET`'s value gets you the full test-agent hat (author + runner scope, sees everything); presenting `ARGUS_RUNNER_TOKEN`'s value gets you the product-agent hat (runner scope only, redacted).docs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L84-86 (under "2. The two kinds of token"))repo/internal/role/role.go#L1-27
    repo/internal/auth/auth.go#L8-9
    role.go maps ARGUS_RUNNER_TOKEN to the product hat (runner-side commands, redacted failure context) and ARGUS_EXECUTOR_SECRET to the test hat with full access (L1-27).
    C-478If the control plane refuses that token, the error now says it was refused (v0.3.48 or later) rather than asking you to log in again.docs.onedroid.ai/argus-quickstart (sources/argus-quickstart.html L133-134 (under "2. The two kinds of token"))
    docs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L98-99 (under "2. The two kinds of token"))
    repo/cmd/argus/doctor_pointer.go#L32-52
    repo/cmd/argus/doctor.go#L290
    doctor_pointer.go detects a control plane refusing the credential and points to that instead of a re-login (L32-52); PARTIAL: the v0.3.48 tag is not checkable.
    C-480`ARGUS_RUNNER_TOKEN` is minted per execution-plane instance, not per persondocs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L145 (under "2. The two kinds of token"))repo/cmd/argus/runnerid_cmd.go#L23
    repo/cmd/argus/runnerid_cmd.go#L39-69
    `argus runner-id mint` requires --instance-id and binds the minted runner id to that instance (runnerid_cmd.go L23, L39-69).
    C-482From v0.3.52 `preflight` then also checks that the Docker daemon answers and that the host ports the shared observability stack publishes are freedocs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L208-210 (under "4. Prove read access before wiring anything else"))repo/internal/preflight/preflight.go#L127
    repo/internal/preflight/preflight.go#L363-412
    preflight checks the Docker daemon (L363-381) and that the host ports the compose observability stack publishes are free (L393-412); PARTIAL: the v0.3.52 tag is not checkable.
    C-483A working cloud tester no longer fails. If `ARGUS_TOKEN` holds a control-plane token and no local role map is set, `local-hats` is a warning, not a failure.docs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L246-247 (under "4. Prove read access before wiring anything else"))repo/internal/doctor/hats.go#L80-85
    CheckLocalHats returns StatusWarn, not fail, when ARGUS_TOKEN holds a control-plane token and no local role map is set ('A working cloud tester', hats.go L80-85).
    C-484This repository is the open part of OneDroid Argusraw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L3 (under "OneDroid Argus runner"))sources/github-repo.json#L1
    repo/LICENSE#L1
    sources/argus-runner-README.md#L3
    The repository is public (visibility public, private false) under the MIT LICENSE and holds the runner and CLI while `serve --mode control` is refused (main.go L2906).
    C-485OneDroid Argus is an end-to-end test and certification service.raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L7 (under "OneDroid Argus runner"))repo/internal/argus/chain_scenario.go#L446
    repo/internal/certificate/verify.go#L3
    repo/docs/CERTIFICATE.md#L10-19
    The open runner executes end-to-end scenarios and ships the certificate format and verifier, which would not exist without a test and certification service; the hosted service itself is not public.
    C-486The control plane signs that statement and anchors it on a public chain.raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L19-20 (under "What a OneDroid Argus certificate proves today, and what it does not"))repo/internal/certificate/verify.go#L9-10
    repo/internal/certificate/verify.go#L60-63
    https://base-sepolia.blockscout.com/api/v2/transactions/0x53ff7fd12710b90183bb166821766fcc5b6152b7e7b05d1a08161575f4632228#fetched/001.txt L1 status ok block 48212445
    verify.go reads each anchor back under its declared allowed signer, and a fetched Base Sepolia transaction carries a JSON anchor payload with status ok.
    C-488`argus certificate verify` reports `NOT ANCHORED` for fields a given certificate format does not put on chain.raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L36-37 (under "What a OneDroid Argus certificate proves today, and what it does not"))repo/internal/certificate/verify.go#L214
    repo/internal/certificate/certificate.go#L21
    repo/docs/CERTIFICATE.md#L51
    `argus certificate verify` emits NOT ANCHORED for fields the certificate format does not anchor (verify.go L214, certificate.go L21).
    C-489Argus can also run under Docker Compose with no cluster, but that path needs the onboarding kit, which this repository does not shipraw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L45-46 (under "Quickstart"))repo/README.md#L99-100
    repo (ba724fb1) git ls-tree onboarding/: HOW-TO-ARGUS-CONFIG.md, argus-config.template.yaml only
    repo/cmd/argus/preflight_compose.go#L16
    The compose path exists in code (preflight_compose.go) and the onboarding/ directory holds only two docs, no onboard.sh or compose scripts.
    C-490`verify` never contacts the control plane.raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L110 (under "Verify a certificate"))repo/internal/certificate/verify.go#L3-4
    repo/internal/certificate/verify.go#L15-29
    verify.go is offline and independent of the control plane: its imports are the stdlib without net/http plus chainread, ots and artifactmeasure (L15-29).
    C-491After a certified run, its author can reveal the sealed scenarios. The reveal document holds every sealed scenario, the Merkle proofs that tie each one to the set anchored before the run, and the run's anchors.raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L119-120 (under "Check a reveal, and replay a revealed scenario"))repo/internal/reveal/document.go#L3
    repo/internal/reveal/document.go#L32-37
    repo/internal/reveal/document.go#L67-84
    The reveal Document type holds the sealed scenario files, a Merkle root with per-file inclusion proofs, and the run's anchors[] (document.go L3, L32-37, L67-84); PARTIAL: that only the author can reveal is control-plane behaviour.
    C-492`replay` runs one revealed scenario against a deployment you run yourself and compares your outcome with the verdict the reveal declares.raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L131-132 (under "Check a reveal, and replay a revealed scenario"))repo/internal/reveal/replay.go#L9-17
    repo/internal/reveal/replay.go#L50-76
    Replay runs one revealed scenario through the caller's own executor and compares the observed outcome with the reveal's declared verdict (replay.go L50-76).
    C-493`--against` must be the artefact digest the reveal certified (it is refused otherwise), so you replay the same build.raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L132-133 (under "Check a reveal, and replay a revealed scenario"))repo/internal/reveal/replay.go#L21-22
    repo/internal/reveal/replay.go#L47-53
    repo/cmd/argus/reveal_cmd.go#L38
    Replay refuses with exit 2 when --against does not equal the reveal's declared ArtifactDigest (replay.go L47-53).
    C-494Anchors on a private chain (for example `argus-dev-poa`) read as UNREACHABLE from outside: the public check is Base Sepolia and Bitcoin.raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L133-134 (under "Check a reveal, and replay a revealed scenario"))repo/internal/reveal/verify_anchors.go#L36-50
    repo/internal/chainread/ots/headers.go#L24
    repo/README.md#L134
    Any anchor read that fails without a configured independent client is reported UNREACHABLE (verify_anchors.go L36-50) and Bitcoin is checked through public header sources (ots/headers.go L24); PARTIAL: the private chain's unreachability from outside is inferred from that rule.
    C-495`internal/reveal` | the reading side of a reveal document: `anchor verify` and `replay` (it assembles nothing and writes no chain)raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L143 (under "What is here"))repo/internal/reveal/document.go#L8-9
    repo/internal/reveal/verify_anchors.go#L34-36
    internal/reveal reads documents and chain anchors back for `anchor verify` and `replay`, and states it does not assemble a reveal or write a ledger or chain (document.go L8-9).
    C-496The executor does sign its own requests to the control plane, with its own instance key.raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L154-155 (under "What is deliberately not here"))repo/internal/federation/identity.go#L74-83
    repo/internal/runner/client.go#L6
    MintJWT signs a JWT bound to the instance id with the instance's ed25519 private key (identity.go L74-83), used by the executor's federation client.
    C-497"description":"OneDroid Argus open runner and tester CLI (MIT). It executes sealed checks against your system, and argus certificate verify checks a certificate against the chain."api.github.com/repos/OneDro1d/argus-runner (sources/github-repo.json L1)sources/github-repo.json#L1
    https://api.github.com/repos/OneDro1d/argus-runner#200
    repo/LICENSE#L1
    The GitHub API answers 200 with this description field, and the repo LICENSE is MIT as the description says.
    C-524Or give your agent onedroid.ai/kit: every step written for it to follow.onedroid.ai/llms-full.txt (sources/llms-full.txt L4599-4600 (PAGE /quickstart; under "Where it leads"))https://onedroid.ai/kit#200
    sources/llms-full.txt#L5591-5672
    onedroid.ai/kit answers 200 and lays out numbered steps for an agent, each with a Check line.
    C-529Open source under Apache-2.0, with no paid tier.onedroid.ai/llms-full.txt (sources/llms-full.txt L5500-5501 (PAGE /kit; under "The method: the Dark Factory kits"))https://api.github.com/repos/OneDro1d/dark-factory#200 license.spdx_id Apache-2.0
    https://github.com/OneDro1d/dark-factory LICENSE (tree path LICENSE)
    The public dark-factory repository reports licence Apache-2.0 and visibility public; PARTIAL: 'no paid tier' is a statement only.
    C-530Product owner, architect, test-first developer and QA, each a skill your agent runs in order.onedroid.ai/llms-full.txt (sources/llms-full.txt L5492-5493 (PAGE /kit; under "The method: the Dark Factory kits"))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/kits/method-core/kit.json#200 skills
    https://onedroid.ai/.well-known/skills/index.json#200
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/docs/DARK-FACTORY-PRIMING.md#200 L88-102
    method-core lists df-product-owner, df-solution-architect, df-tdd-developer and df-qa as skills, and the priming doc's Part 1 pipeline orders the stages.
    C-531Acceptance criteria become a failing test list before any code is written.onedroid.ai/llms-full.txt (sources/llms-full.txt L5493-5494 (PAGE /kit; under "The method: the Dark Factory kits"))https://onedroid.ai/.well-known/skills/df-tdd-developer/SKILL.md#200 L3
    https://onedroid.ai/.well-known/skills/df-tdd-developer/SKILL.md#200 L14-18
    The df-tdd-developer skill makes the test list the PO validation rules and scenarios and starts each case with a failing (RED) test before code.
    C-532An adversary gate checks the evidence for each claim (test output, diffs, exit codes) without redoing the work, and work that fails the gate goes back.onedroid.ai/llms-full.txt (sources/llms-full.txt L5494-5496 (PAGE /kit; under "The method: the Dark Factory kits"))https://onedroid.ai/.well-known/skills/df-adversary-gate/SKILL.md#200 L3
    https://onedroid.ai/.well-known/skills/df-adversary-gate/SKILL.md#200 L12-14
    https://onedroid.ai/.well-known/skills/df-adversary-gate/SKILL.md#200 L66
    The df-adversary-gate skill validates only the presented evidence (exit codes and the like) without redoing the work, and a failed verdict sends the work back to be fixed and re-dispatched.
    C-533Hooks enforce what must not depend on the model choosing to comply: a rule in a prompt is a request, a hook is a control.onedroid.ai/llms-full.txt (sources/llms-full.txt L5496-5497 (PAGE /kit; under "The method: the Dark Factory kits"))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/kits/method-core/kit.json#200 hooks
    https://github.com/OneDro1d/dark-factory hooks/df-stage-gate.py (tree)
    method-core installs hooks such as df-stage-gate.py from the public repo; PARTIAL: the prompt-versus-control framing is an opinion, not a checkable fact.
    C-539For ChatGPT, Gemini or a plain API model, paste DARK-FACTORY-PRIMING.md into the conversation.onedroid.ai/llms-full.txt (sources/llms-full.txt L4571-4573 (PAGE /quickstart; under "No skill system? One page"))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L5683-5685 (PAGE /kit; under "No skill system?"))
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/docs/DARK-FACTORY-PRIMING.md#200 L1
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/docs/DARK-FACTORY-PRIMING.md#200 L538
    DARK-FACTORY-PRIMING.md is public (200), titled a self-contained priming doc for skill-less agents, with Part 5 adaptation notes for ChatGPT, Gemini and plain API models.
    C-540Where it stops for you: your sign-ups and logins (OneDroid Argus, Synapse, Engram and GitHub), which kit, the name of your repo, and the tools you connect.onedroid.ai/llms-full.txt (sources/llms-full.txt L5504-5505 (PAGE /kit; under "The method: the Dark Factory kits"))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/START-HERE.md#200 L62
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/START-HERE.md#200 L79-80
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/START-HERE.md#200 L198
    The runbook marks HUMAN at the gh login, the kit, name and owner choice, and connecting tools; PARTIAL: the Argus, Synapse and Engram sign-up stops were not traced.
    C-542The runbook is START-HERE.md in the public repository.onedroid.ai/llms-full.txt (sources/llms-full.txt L5591-5593 (PAGE /kit; under "Part 4 · The method: the Dark Factory kits"))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/START-HERE.md#200
    https://api.github.com/repos/OneDro1d/dark-factory#200 visibility public
    START-HERE.md exists at starter-kit/instance/START-HERE.md in the public dark-factory repository.
    C-543Every step ends with a check: run it, and do not move on until it passes.onedroid.ai/llms-full.txt (sources/llms-full.txt L5512-5513 (PAGE /kit; under "The steps, in the same order, each with its check."))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/START-HERE.md#200 L7
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/START-HERE.md#200 L95,L120,L160,L194
    The runbook's steps each end in a **Check:** line, and L7 says not to move on until it passes.
    C-544Stop and ask the human at the steps marked HUMAN, and whenever a runbook says to stop (a setup that is somebody else's, a tool that is missing).onedroid.ai/llms-full.txt (sources/llms-full.txt L5513-5514 (PAGE /kit; under "The steps, in the same order, each with its check."))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/START-HERE.md#200 L7-8
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/START-HERE.md#200 L45
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/START-HERE.md#200 L61
    The runbook marks HUMAN steps and tells the agent to stop for somebody else's setup (L45) and for a missing tool (L61).
    C-546You need `git`, `jq`, `bash`, `python3`, `gh` and `claude`, and a `gh` login that can see the repository.onedroid.ai/llms-full.txt (sources/llms-full.txt L5599-5601 (PAGE /kit; under "Check the tools"))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/START-HERE.md#200 L53
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/bootstrap.sh#200 L85
    The runbook's tool check loops over git jq bash python3 gh claude (L53) and bootstrap.sh refuses --kit without python3 (L85).
    C-547Seven kits today: `method-core` (the method itself), `dev` (writing and shipping code), `knowledge-worker` (documents rather than code), `agent-ops`, `code-review`, `distributed-systems`, `frontend`.onedroid.ai/llms-full.txt (sources/llms-full.txt L5614-5617 (PAGE /kit; under "Clone the method and list the kits"))https://api.github.com/repos/OneDro1d/dark-factory/git/trees/main?recursive=1#200 kits/*/kit.json
    The public tree holds exactly seven kits/*/kit.json: agent-ops, code-review, dev, distributed-systems, frontend, knowledge-worker, method-core.
    C-548The bootstrap writes a new directory beside the checkout, pinned to today's commit, and the copy goes to a private repository.onedroid.ai/llms-full.txt (sources/llms-full.txt L5629-5631 (PAGE /kit; under "Make the human's own copy"))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/bootstrap.sh#200 L16
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/bootstrap.sh#200 L98-102
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/START-HERE.md#200 L88
    bootstrap.sh defaults the target to a sibling ../<name> and pins the Tier-1 commit read from the remote's main at that moment, and the runbook pushes it to a --private repo.
    C-549Every machine gets its own record under `instances/`, named as START-HERE.md step 3 says.onedroid.ai/llms-full.txt (sources/llms-full.txt L5648-5649 (PAGE /kit; under "Add this machine and install"))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/START-HERE.md#200 L126-145
    The runbook's step 3 writes instances/$M/loom.lock.json for each machine.
    C-550`0` means installed and proven, `1` means a precondition failed and nothing was installed, `2` means the result does not match the record.onedroid.ai/llms-full.txt (sources/llms-full.txt L5653-5655 (PAGE /kit; under "Add this machine and install"))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/install.sh#200 L38-40
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/install.sh#200 L826
    install.sh documents exit 0 installed and LOCKED, 1 a precondition failed, 2 installed but the result does not match the lockfile; PARTIAL: 0 means LOCKED, and a missing prove.sh only warns UNPROVEN (L826), so 'proven' is not guaranteed by 0.
    C-554A fresh session exercises every gate the kit installs and commits its report to the human's repo by itself.onedroid.ai/llms-full.txt (sources/llms-full.txt L5672-5673 (PAGE /kit; under "Validate"))https://api.github.com/repos/OneDro1d/dark-factory/git/trees/main?recursive=1#200 boot-kit/scripts/validate.sh
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/START-HERE.md#200 L257-261
    The public repo ships boot-kit/scripts/validate.sh, which the runbook says writes and commits VALIDATE-REPORT-*.md; PARTIAL: that it exercises every gate the kit installs was not traced in the script.
    C-555It inlines every stage skill and the adversary gate.onedroid.ai/llms-full.txt (sources/llms-full.txt L5685 (PAGE /kit; under "No skill system?"))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/docs/DARK-FACTORY-PRIMING.md#200 L230-479
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/docs/DARK-FACTORY-PRIMING.md#200 L147
    The priming doc inlines stages 1-7 (Part 3) and the adversary gate (Part 2.2, Stage 6).
    C-556Each skill is also served from this site: the skills index lists them, and each lives at `/.well-known/skills/<name>/SKILL.md`.onedroid.ai/llms-full.txt (sources/llms-full.txt L5689-5691 (PAGE /kit; under "Only the skills?"))https://onedroid.ai/.well-known/skills/index.json#200 application/json
    https://onedroid.ai/.well-known/skills/df-adversary-gate/SKILL.md#200 text/markdown
    The skills index answers 200 listing 43 skills, and a listed skill is served at /.well-known/skills/<name>/SKILL.md.
    C-558START-HERE.md ends with a table of the failures people actually hit.onedroid.ai/llms-full.txt (sources/llms-full.txt L5696 (PAGE /kit; under "Something went wrong?"))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/START-HERE.md#200 L328-342 ("## When it goes wrong" symptom table, last lines of a 342-line file)
    https://api.github.com/repos/OneDro1d/dark-factory#200 ("private":false)
    sources/llms-full.txt#L5696
    The public START-HERE.md ends with a symptom/cause table (L328-342); PARTIAL: that these are the failures people actually hit is not evidenced.
    C-564A run still needs an account there, available by request at the time (sign-up opened later; see /argus).onedroid.ai/llms-full.txt (sources/llms-full.txt L6258-6259 (PAGE /changelog; under "The OneDroid Argus runner is open source"))
    onedroid.ai/changelog.md (sources/changelog.md L149-150 (under "The OneDroid Argus runner is open source"))
    repo/README.md#L41-43 (You need an account on the hosted control plane at https://argus-dev.onedroid.ai)
    https://argus-dev.onedroid.ai/#200 ("Argus MCP — Control Plane")
    https://argus-dev.onedroid.ai/assets/index-8XgKDqxK.js#200 (signUpForceRedirectUrl)
    The runner README makes a control-plane account a precondition and the live app wires a sign-up flow; PARTIAL: that accounts were by request only at the time is a historical statement with no artefact.
    C-565OneDroid Synapse is a governed MCP gateway: your agents connect to one URL, and every tool call is authenticated, policy-checked, and logged while the hub's activity logging is on (fail-open: a logging failure never blocks a call) to an audit log (called the event log in the product).onedroid.ai/llms-full.txt (sources/llms-full.txt L6616-6618 (PAGE /terms; under "2. The Service"))https://synapse.onedroid.ai/agent/mcp#401 ({"code":"missing_token"})
    https://synapse.onedroid.ai/assets/index-zeCwIjse.js#200 (activity_logging, Export CSV, /tools/toggles)
    sources/llms-full.txt#L6616-6618
    The one gateway URL refuses an unauthenticated call and the app exposes an activity-logging switch and log export; PARTIAL: per-call policy checks and fail-open logging are not shown, and repo greps for policy-checked/fail-open are not-found.
    C-566The kit to build a dark software factory, in four parts ordered by what gives the most for the effort: testing first (OneDroid Argus), then eyes and hands (OneDroid Synapse), memory and context (OneDroid Engram), and the open-source Dark Factory kits as the method that ties them together. For the human, why each part is needed and what it does; for the coding agent, every step with its check, ending with a validation report committed in the user's own private repository. Every link in one place.onedroid.ai/llms.txt (sources/llms.txt L69 (under "Docs"))sources/llms-full.txt#L5390-5725 (PAGE /kit: four parts in order, steps 01-10 each with Check, "Everything in one place.")
    https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/START-HERE.md#200 L250-258 (validate.sh commits and pushes VALIDATE-REPORT to the human's repo)
    sources/llms.txt#L69
    The /kit page it describes has the four parts in that order, a check per agent step, a private-repo step (gh repo create --private) and a link section, and the public runbook's validate step commits the report.
    C-567a builder uses a builder token and never signs indocs.onedroid.ai/argus-quickstart (sources/argus-quickstart.html L189 (under "3. Point the CLI at your control plane"))
    docs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L168 (under "3. Point the CLI at your control plane"))
    repo/internal/role/role.go#L1-8 (the bearer token IS the hat: ARGUS_RUNNER_TOKEN -> product)
    repo/README.md#L59 (cloud-login --scope author for the tester)
    sources/argus-quickstart.md#L168
    In code the builder (product) hat comes from a bearer token, and sign-in is documented for author scope; PARTIAL: that no builder path ever needs a sign-in is not shown exhaustively.
    C-568Without `--scope` the sign-in stops before it startsdocs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L171 (under "3. Point the CLI at your control plane"))repo/cmd/argus/main.go#L1495 ("cloud-login: --scope is required, there is no default")
    sources/argus-quickstart.md#L171-176
    cloud-login returns a usage error with the quoted message when --scope is missing, before any sign-in.
    C-569OneDroid Argus tests the running system against what its architecture and its owners say it does, with checks sealed before the build.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L247-248 (under "Your architecture says what may talk to what. Does the running system agree?"))https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#fetched/005.txt (set_hash, holdout_root, scenario_count 7, 2026-10-07T19:58:34Z)
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#fetched/006.txt (verdict failed, 2026-10-07T19:59:26Z)
    repo/README.md#L7-10
    An anchored sealed set precedes an anchored verdict and the open runner executes checks against the system; PARTIAL: the record shows sealing before the run, not before the build, and testing against what owners say is a framing.
    C-571Every figure on it links to the page it comes from on onedroid.ai.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L258 (under "Your architecture says what may talk to what. Does the running system agree?"))sources/partners-financial-services.html#L398 (23% and 35% linked to /argus)
    sources/partners-financial-services.html#L402-406 ($1.01/$1.97/$5.30 paragraph linked to /case-study-claude-code-alone-vs-tester)
    sources/partners-financial-services.html#L371 (7/5/2/0 table followed by /case-study-finos-calm)
    The page's figures carry links to the onedroid.ai pages holding them; PARTIAL: the "50,000 lines overnight" figure (L285) and the results table have no link on the figure itself.
    C-574The set is sealed, and its hash goes on a cryptographic ledger before the work is judged.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L296-297 (under "02What OneDroid Argus does"))https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#fetched/005.txt (block 47817413, payload set_hash and holdout_root, 19:58:34Z)
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#fetched/006.txt (block 47817439, verdict, 19:59:26Z)
    A public chain transaction carries the sealed set's hash at a block before the verdict transaction.
    C-575The builder holds a credential that can ask for a run and read the verdict, and cannot read a scenario or an expected value.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L298-299 (under "02What OneDroid Argus does"))repo/internal/role/role.go#L44-56 (product hat: CanAccessScenarios false, MustRedactExpected, MustOmitScenario)
    repo/internal/runner/limitedreport.go#L64 (keep list includes layers[].scenarios[].failure.expected)
    The role policy bars the builder hat from scenarios and redacts expected values in its failure bundle; PARTIAL: the builder's limited report keep list still sends failure.expected, so the expected-value clause is not backed.
    C-576It drives HTTP, the message broker, the database, MCP tool calls, web pages in a real browser, permissions and rate limits.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L304-305 (under "02What OneDroid Argus does"))repo/README.md#L7-8
    repo/internal/scenario/amqpstep.go
    repo/internal/scenario/dbstep.go#L16
    The open runner has step engines for HTTP, AMQP, database, MCP and web UI, plus Permissions and Rate Limiting layers.
    C-577Checks can be generated from a CALM architecture file with argus calm import, and written by hand for the paths the file does not describe.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L305-307 (under "02What OneDroid Argus does"))repo/internal/calm/import.go#L829
    repo/cmd/argus/calm_cmd_test.go#L10
    sources/case-study-finos-calm.md#L63
    argus calm import is implemented in the open repo, and the case study's setup shows six generated checks plus one by hand.
    C-579When the tester reveals the set, any scenario can be replayed by the runner against the same artefact: by you, by your auditor, by anyone.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L312-313 (under "02What OneDroid Argus does"))repo/README.md#L117-134 (argus replay --reveal ... --against sha256:<digest>, without a OneDroid account)
    The open CLI implements replay of a revealed scenario against the certified artefact digest, refused otherwise.
    C-580Database access is read-only, except a clean-up a scenario declares, run with the credential you give it.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L317 (under "02What OneDroid Argus does"))repo/internal/scenario/dbstep.go#L16 (one read-only SELECT)
    repo/internal/scenario/cleanup.go#L35-42 (## CLEANUP sql runs against targets.database)
    Database steps are read-only SELECTs and a scenario-declared CLEANUP block is the write path against the configured database target.
    C-581Nothing can be made to fail on purpose unless the system's operator has listed it, and a load ramp never targets a live broker.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L317-319 (under "02What OneDroid Argus does"))repo/docs/fault-steps.md#L8 (No fault_allowed_targets block = no fault step can run)
    repo/internal/config/faultallowed.go#L92
    repo/internal/config/loadallowed.go#L151 (only if ..., never a live one)
    Fault steps need an operator allowlist and the load allowlist refuses targets that look live or shared.
    C-584Our team ran one sealed set of checks with OneDroid Argus against the FINOS CALM "trades API and MCP" demo: a trades REST API, an MCP server for an AI agent with a guardrail on restricted symbols, and an A2A server.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L331-333 (under "03The proof: a FINOS CALM trading demo"))https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#fetched/005.txt (scenario_count 7)
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#fetched/006.txt (artifact_digest sha256:feb820a3..., tallies passed 5 failed 2 errored 0)
    sources/case-study-finos-calm.md#L55-57
    One sealed set of 7 and its verdict are on chain, and the bound digest matches the trades-mcp-server image the case study declares; PARTIAL: the demo's composition (REST API, guarded MCP server, A2A server) is stated only.
    C-585They were sealed on 7 October 2026, before the run.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L334-335 (under "03The proof: a FINOS CALM trading demo"))https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#fetched/005.txt (timestamp 2026-10-07T19:58:34Z)
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#fetched/006.txt (timestamp 2026-10-07T19:59:26Z)
    The seal transaction is dated 7 October 2026 and precedes the verdict transaction.
    C-588A check generated from the architecture file asks that question on every deployment.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L355-356 (under "03The proof: a FINOS CALM trading demo"))repo/internal/calm/import.go#L490-498 ("CALM must not connect" check with an unreachable step)
    The CALM importer generates a must-not-connect check from the architecture file, which runs with each run of the set.
    C-589Before the run, OneDroid Argus read the image digests actually running and found exactly the ones declared.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L357-358 (under "03The proof: a FINOS CALM trading demo"))repo/internal/artifactmeasure/measure.go#L203
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#fetched/006.txt (artifact_digest sha256:feb820a3...)
    sources/case-study-finos-calm.md#L57
    The open code reads running digests and refuses a mismatch, and the anchored verdict binds the declared MCP-server digest; PARTIAL: that all three running digests matched is in the private bundle only.
    C-590A run that declared a digest that was not running was refused before any check ran.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L358-359 (under "03The proof: a FINOS CALM trading demo"))repo/internal/artifactmeasure/measure.go#L203 ("is not among the image digests running in the system under test — no scenario was run and nothing was bound")
    sources/case-study-finos-calm.md#L137
    The refusal message quoted by the case study is the one in the open code; PARTIAL: the refused run itself is in the private bundle only.
    C-598Sealed checks that declared connections answer and undeclared ones are refused, on the running systemonedroid.ai/partners-financial-services (sources/partners-financial-services.html L385 (under "04Where it fits next to the FINOS SDLC Common Controls"))repo/internal/calm/import.go#L490-498
    sources/case-study-finos-calm.md#L75-77
    The CALM importer makes checks that declared connections answer and undeclared ones are unreachable, run against the deployed system.
    C-599The images running are read by digest and must equal the ones declared before any check runsonedroid.ai/partners-financial-services (sources/partners-financial-services.html L386 (under "04Where it fits next to the FINOS SDLC Common Controls"))repo/internal/artifactmeasure/measure.go#L203
    The open code refuses a run whose declared digest is not among the running image digests, before any scenario runs.
    C-600A verdict bound to one build, with the sealed set and the verdict recorded on a ledger and a certificate anyone can checkonedroid.ai/partners-financial-services (sources/partners-financial-services.html L387 (under "04Where it fits next to the FINOS SDLC Common Controls"))https://base-sepolia.blockscout.com/api/v2/transactions/0x9fccbd30504e7e31554455dfc27bbf27ca5cba4cd5e30d9f607a6355aea32adc#fetched/005.txt (set_hash)
    https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#fetched/006.txt (run_id, artifact_digest, verdict)
    repo/cmd/argus/certificate_cmd.go#L9-11
    The sealed set and a verdict bound to one artefact digest are on a public chain and the open CLI verifies certificates against them.
    C-601A sealed verdict about one build, a record your release process can keep and requireonedroid.ai/partners-financial-services (sources/partners-financial-services.html L388 (under "04Where it fits next to the FINOS SDLC Common Controls"))https://base-sepolia.blockscout.com/api/v2/transactions/0x8718c0b64c7fd35feb06029e1af1574b944d35c2afde6676a114132b7353ce7a#fetched/006.txt (verdict bound to artifact_digest)
    repo/README.md#L18-23
    An anchored verdict bound to one build exists as a keepable record; PARTIAL: that a release process can require it is the customer's integration and is stated only.
    C-606The runner is free and open source.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L408 (under "05What it costs"))repo/LICENSE#L1 (MIT License)
    repo/README.md#L12
    The runner repository carries an MIT LICENSE.
    C-607The hosted control plane, which seals the withheld scenarios and adjudicates the verdict, is the priced part, with sign-up at argus-dev.onedroid.aionedroid.ai/partners-financial-services (sources/partners-financial-services.html L408-409 (under "05What it costs"))https://argus-dev.onedroid.ai/#200 ("Argus MCP — Control Plane")
    https://argus-dev.onedroid.ai/assets/index-8XgKDqxK.js#200 (signUpForceRedirectUrl)
    sources/pricing.md#L139
    The hosted control plane answers at argus-dev with a sign-up flow wired in; PARTIAL: that it is the priced part is stated on /pricing with no published price.
    C-609Not a score. Passed, failed, errored or not_applicable, per scenario, with the evidence.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L422 (under "06What it is not"))repo/internal/certificate/verify.go#L163-176 (passed, failed, errored, degraded, not_applicable buckets)
    repo/internal/runner/limitedreport.go#L62 (layers[].scenarios[].status per scenario)
    Outcomes are per-scenario statuses in four (plus degraded) buckets; PARTIAL: "not a score" is a negative with no artefact.
    C-612It runs inside your environment and its source is open under MITonedroid.ai/partners-financial-services (sources/partners-financial-services.html L438-439 (under "07How to start"))repo/LICENSE#L1 (MIT License)
    repo/README.md#L7-8 (executes them next to your system)
    repo/README.md#L80-81 (render-k8s --sut-namespace into your namespace)
    The runner source is MIT and deploys into the user's own cluster namespace.
    C-613A run needs an account on the hosted control plane: anyone can create one at argus-dev.onedroid.ai (e-mail, Google or Microsoft).onedroid.ai/partners-financial-services (sources/partners-financial-services.html L441 (under "07How to start"))https://argus-dev.onedroid.ai/#200
    https://argus-dev.onedroid.ai/assets/index-8XgKDqxK.js#200 (Clerk pk_test key, signUpForceRedirectUrl)
    repo/README.md#L41-43
    The control plane answers 200 with a Clerk sign-up flow; PARTIAL: the provider list (e-mail, Google, Microsoft) cannot be read from allowed hosts, and fetched/008 (clerk.onedroid.ai, oauth_microsoft enabled:false) is a different Clerk instance from argus-dev's.
    C-617OneDroid Argus — built by Providentia Worldwide.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L464 (under "07How to start"))repo/LICENSE#L3 (Copyright (c) 2026 Providentia Worldwide)
    sources/partners-financial-services.html#L464
    The runner's LICENSE names Providentia Worldwide as copyright holder.
    C-618Every scenario ends as passed, failed, errored or not_applicable (it could not run; a load run has one more outcome, degraded), and neither an error nor not_applicable is ever a pass.onedroid.ai/llms-full.txt (sources/llms-full.txt L307 (PAGE /; under "Does OneDroid Argus score my build?"))repo/docs/CERTIFICATE.md#L69-74
    repo/docs/CERTIFICATE.md#L139-140 (passed means every scenario passed or was not applicable, with at least one passed)
    repo/internal/certificate/verify.go#L184-189
    The certificate format has the outcome buckets plus degraded, and the verifier refuses a pass with an errored scenario or with every scenario not_applicable.
    C-620Every tool call is written to an event log you can query and export while the hub's activity logging is on (fail-open: a logging failure never blocks a call).onedroid.ai/llms-full.txt (sources/llms-full.txt L4712 (PAGE /security; under "Append-only audit log"))
    onedroid.ai/security.md (sources/security.md L103 (under "Append-only audit log"))
    https://synapse.onedroid.ai/assets/index-zeCwIjse.js#200 (activity_logging, Export CSV)
    sources/security.md#L103
    The Synapse app exposes an activity-logging switch and a CSV export; PARTIAL: fail-open behaviour and every call being logged are not shown (repo greps for fail-open not-found).
    C-621The kit's install step runs through the `claude` CLI today; any harness that reads SKILL.md can follow the method.onedroid.ai/llms-full.txt (sources/llms-full.txt L5374 (PAGE /dark-factory; under "Harness-agnostic"))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/START-HERE.md#200 L53-55 (tool check requires claude, claude -p)
    https://api.github.com/repos/OneDro1d/dark-factory/git/trees/main?recursive=1#200 (docs/DARK-FACTORY-PRIMING.md)
    The public runbook's install path requires the claude CLI; PARTIAL: that any SKILL.md-reading harness can follow the method is not shown.
    C-622every invocation is written to the append-only audit log while the hub's activity logging is ononedroid.ai/llms-full.txt (sources/llms-full.txt L6563-6564 (PAGE /privacy; under "Security"))
    onedroid.ai/privacy.md (sources/privacy.md L172-173 (under "Security"))
    https://synapse.onedroid.ai/assets/index-zeCwIjse.js#200 (activity_logging)
    sources/privacy.md#L172-173
    The app exposes the activity-logging switch the clause depends on; PARTIAL: append-only storage and per-invocation writes are not shown (repo greps for append-only not-found).
    C-624Automated end-to-end testing for AI-written code.onedroid.ai/llms-full.txt (sources/llms-full.txt L283 (PAGE /; under "What is OneDroid Argus?"))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L5431 (PAGE /kit; under "Testing: OneDroid Argus"))
    repo/README.md#L7-8 (an end-to-end test and certification service; the runner executes them next to your system)
    sources/llms-full.txt#L283
    The open runner is an end-to-end test executor; PARTIAL: "for AI-written code" is positioning, not a property of the artefact.
    C-625what to trigger, where to look, what a healthy answer is.onedroid.ai/llms-full.txt (sources/llms-full.txt L2779 (PAGE /argus; under "Plain scenarios, not test scripts."))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L5433-5434 (PAGE /kit; under "Testing: OneDroid Argus"))
    onedroid.ai/argus.md (sources/argus.md L43 (under "Plain scenarios, not test scripts."))
    repo/examples/http-hello/scenarios (## TRIGGER L8, ## VERIFY L11, ## EXPECT L14)
    repo/internal/scenario/sections.go#L113
    The scenario format has TRIGGER, VERIFY and EXPECT sections, matching what to trigger, where to look and what a healthy answer is.
    C-628The kit to build a dark software factory with AI agents, in the order that gives you the most: testing with OneDroid Argus, eyes and hands with OneDroid Synapse, memory and context with OneDroid Engram, and the open-source Dark Factory kits as the method that ties them together.onedroid.ai/llms-full.txt (sources/llms-full.txt L5394 (PAGE /kit))repo/LICENSE#L1 (MIT runner)
    https://api.github.com/repos/OneDro1d/dark-factory#200 (visibility public, license apache-2.0)
    https://synapse.onedroid.ai/#200
    The kit's parts exist publicly (MIT Argus runner, public Apache-2.0 Dark Factory repo, live Synapse app); PARTIAL: the order that gives the most is the publisher's judgement.
    C-632The checks run against the running system with nothing mocked.onedroid.ai/llms-full.txt (sources/llms-full.txt L5434-5435 (PAGE /kit; under "Testing: OneDroid Argus"))repo/README.md#L7-8 (the runner executes them next to your system: HTTP, database state, message flow, log lines, load, web UI)
    The open runner drives the declared live targets; PARTIAL: whether anything is mocked depends on the targets the user configures.
    C-633Your coding agent's credential can ask for a run and read the verdict, and cannot read a check or an expected value.onedroid.ai/llms-full.txt (sources/llms-full.txt L5435-5436 (PAGE /kit; under "Testing: OneDroid Argus"))repo/internal/role/role.go#L44-56
    repo/internal/runner/limitedreport.go#L64 (keep list includes layers[].scenarios[].failure.expected)
    The builder (product) hat cannot list or read scenarios; PARTIAL: the limited report sent to the builder keeps failure.expected, so the expected-value clause is not backed.
    C-635The runner is open source under MIT; the hosted control plane is open to sign up and evaluate.onedroid.ai/llms-full.txt (sources/llms-full.txt L5443-5444 (PAGE /kit; under "Testing: OneDroid Argus"))repo/LICENSE#L1 (MIT License)
    https://argus-dev.onedroid.ai/assets/index-8XgKDqxK.js#200 (signUpForceRedirectUrl)
    The runner is MIT and the hosted control plane serves a sign-up-wired app; PARTIAL: that sign-up is open to anyone is not shown from allowed hosts.
    C-636A gateway between your agents and your accounts.onedroid.ai/llms-full.txt (sources/llms-full.txt L5456 (PAGE /kit; under "Eyes and hands: OneDroid Synapse"))https://synapse.onedroid.ai/agent/mcp#401 ({"code":"missing_token"})
    https://synapse.onedroid.ai/assets/index-zeCwIjse.js#200 (Google Workspace, Atlassian, Slack)
    A live gateway endpoint for agents exists, with account connectors in the app.
    C-637You connect the services you choose (Google Workspace, Atlassian, Slack and others), and your agents connect once.onedroid.ai/llms-full.txt (sources/llms-full.txt L5457-5458 (PAGE /kit; under "Eyes and hands: OneDroid Synapse"))https://synapse.onedroid.ai/assets/index-zeCwIjse.js#200 (Google Workspace, Atlassian, Slack)
    https://synapse.onedroid.ai/agent/mcp#401
    The app lists the named connectors and agents connect to a single /agent/mcp URL.
    C-644An autonomy policy says what the agents decide alone, what they log, and what always stops for a human.onedroid.ai/llms-full.txt (sources/llms-full.txt L5497-5498 (PAGE /kit; under "The method: the Dark Factory kits"))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/skills/work-autonomously/SKILL.md#200 L28 (default is action), L137 (log what was ambiguous ... and what you decided), L140 (Hard stops — stop and ask before)
    The public kit ships an autonomy skill that defines what agents do alone, what they log and the hard stops that wait for a human.
    C-647Without access to a release, build the CLI from the open-source runneronedroid.ai/llms-full.txt (sources/llms-full.txt L5529-5530 (PAGE /kit; under "Get an account and the CLI"))repo/BUILDING.md#L17-21
    sources/github-repo.json#L1 private=false license=mit
    repo/cmd/argus/main.go#L749
    The public MIT runner repository has a cmd/argus entry point and BUILDING.md gives the same 'go build -o argus ./cmd/argus' then './argus version' steps.
    C-648Check: `argus version` prints its version. It needs no token.onedroid.ai/llms-full.txt (sources/llms-full.txt L5537 (PAGE /kit; under "Get an account and the CLI"))repo/cmd/argus/main.go#L745-751
    repo/BUILDING.md#L20
    main.go handles cmd == 'version' above the token gate on purpose ('Demanding a token to be told a build number would make the version check useless') and emits the build version.
    C-650The tester's credential writes, seals and reveals; the builder's can start a run and read the verdict, never a scenario or an expected value.onedroid.ai/llms-full.txt (sources/llms-full.txt L5543-5545 (PAGE /kit; under "Keep the tester and the builder apart"))repo/internal/role/role.go#L2-4
    repo/internal/role/role.go#L52
    repo/internal/role/role.go#L56
    role.go makes the token the hat (ARGUS_RUNNER_TOKEN product, ARGUS_EXECUTOR_SECRET test) and the product hat must redact expected values and omit scenarios; PARTIAL: the tester-side write/seal/reveal grants live in the closed control plane and are not shown.
    C-651Check: the tester and the builder use different credentials, and the coding agent holds only the builder's.onedroid.ai/llms-full.txt (sources/llms-full.txt L5549-5550 (PAGE /kit; under "Keep the tester and the builder apart"))repo/internal/role/role.go#L2-4
    sources/llms-full.txt#L5549-5550
    The runner code assigns the role from two distinct tokens, so different credentials per role is implemented; PARTIAL: that the coding agent holds only the builder's is a setup step the reader performs, not shown by any artefact.
    C-655Through the Synapse hub, where the Engram connection is already on, or HUMAN: sign in at engram.onedroid.ai and create a library token for `https://engram.onedroid.ai/mcp`.onedroid.ai/llms-full.txt (sources/llms-full.txt L5580-5582 (PAGE /kit; under "Write one thing, search it back"))https://engram.onedroid.ai#200
    https://engram.onedroid.ai/mcp#401
    sources/llms-full.txt#L5580-5582
    The Engram host answers 200 and its /mcp endpoint answers 401 without a token, so the named MCP URL exists and demands a token; PARTIAL: library-token creation and the Engram connection being on in the Synapse hub sit behind sign-in and were not checked.
    C-657Execute it top to bottom; the steps below summarise it.onedroid.ai/llms-full.txt (sources/llms-full.txt L5593 (PAGE /kit; under "Part 4 · The method: the Dark Factory kits"))https://raw.githubusercontent.com/OneDro1d/dark-factory/main/starter-kit/instance/START-HERE.md#200
    https://api.github.com/repos/OneDro1d/dark-factory#200
    sources/llms-full.txt#L5592-5593
    The START-HERE.md runbook the page tells the agent to execute exists in the public dark-factory repository; PARTIAL: that the /kit steps faithfully summarise it was not compared.
    C-662Today: Argus Code, automated end-to-end testing of AI-written code with checks the coding agent never sees, in use, its runner open source.onedroid.ai/llms-full.txt (sources/llms-full.txt L36-37 (PAGE /; under "Test the agents where they work: in the workflow, in the fleet, next to the people who sign off."))sources/github-repo.json#L1 private=false license=mit
    repo/LICENSE#L1
    repo/internal/role/role.go#L52
    The runner is public under MIT and the product hat is made to redact expected values and omit scenarios; PARTIAL: 'in use' is not evidenced by any published usage record.
    C-664In a randomised experiment on Facebook and Instagram during the 2020 US election, moving users off the algorithmic feed "substantially decreased the time they spent on the platforms and their activity". Guess and colleagues, Science 381, 2023.onedroid.ai/llms-full.txt (sources/llms-full.txt L52-55 (PAGE /; under "Why we test in situ, and not in a lab."))https://csmapnyu.org/research/academic-research/how-do-social-media-feed-algorithms-affect-attitudes-and-behavior-in-an-election-campaign#fetched/014.txt L1
    sources/llms-full.txt#L52-55
    The NYU CSMaP page quotes the abstract word for word ('Moving users out of algorithmic feeds substantially decreased the time they spent on the platforms and their activity') and cites Guess et al., Science 381, 398-404 (2023), Facebook and Instagram, 2020 US election.
    C-665The three-month effect on polarisation was nullonedroid.ai/llms-full.txt (sources/llms-full.txt L56 (PAGE /; under "Why we test in situ, and not in a lab."))https://csmapnyu.org/research/academic-research/how-do-social-media-feed-algorithms-affect-attitudes-and-behavior-in-an-election-campaign#fetched/014.txt L1
    sources/llms-full.txt#L56
    The abstract says the chronological feed 'did not significantly alter levels of issue polarization, affective polarization ... during the 3-month study period'; 'null' is the page's paraphrase of that.
    C-666The European Commission fined Google EUR 2.42 billion in 2017 for giving "an illegal advantage to another Google product, its comparison shopping service".onedroid.ai/llms-full.txt (sources/llms-full.txt L58-60 (PAGE /; under "Why we test in situ, and not in a lab."))https://ec.europa.eu/commission/presscorner/detail/en/IP_17_1784#fetched/015.body L9
    https://ec.europa.eu/commission/presscorner/detail/en/IP_17_1784#fetched/015.body L14
    sources/llms-full.txt#L58-62
    The EC press release body carries 'Commission fines Google &euro;2.42 billion' and 'giving an illegal advantage to another Google product, its comparison shopping service'; PARTIAL: the 2017 date rests on the release code IP/17/1784, since no date string appears in the saved body (its visible text is only the page title).
    C-667The US Federal Trade Commission's 2023 complaint alleges Amazon biased its search results "to preference Amazon's own products".onedroid.ai/llms-full.txt (sources/llms-full.txt L60-61 (PAGE /; under "Why we test in situ, and not in a lab."))fetched/ftc-webfetch.md#L13-16
    https://www.ftc.gov/news-events/news/press-releases/2023/09/ftc-sues-amazon-illegally-maintaining-monopoly-power#404 (scripted GET, fetched/meta.json)
    The FTC press release, dated September 26, 2023, says 'Biasing Amazon's search results to preference Amazon's own products'; CAVEAT: this rests on the WebFetch quote in ftc-webfetch.md, which has no sha256, because www.ftc.gov answers 404 to scripted clients.
    C-668Perplexity, announcing advertising: "subscriptions alone do not generate enough revenue", with ads "formatted as sponsored follow-up questions", and a promise that "the content of the answers you receive on Perplexity will not be influenced by advertisers".onedroid.ai/llms-full.txt (sources/llms-full.txt L65-67 (PAGE /; under "Why we test in situ, and not in a lab."))https://www.perplexity.ai/hub/blog/why-we-re-experimenting-with-advertising#fetched/017.txt L1
    sources/llms-full.txt#L65-68
    Perplexity's Nov 12, 2024 post contains all three quotes verbatim: 'the content of the answers you receive on Perplexity will not be influenced by advertisers', 'formatted as sponsored follow-up questions' and 'subscriptions alone do not generate enough revenue'.
    C-669OpenAI announced ads in ChatGPT for free and Go users in January 2026, with a similar promise (SiliconANGLE, 16 January 2026).onedroid.ai/llms-full.txt (sources/llms-full.txt L69-70 (PAGE /; under "Why we test in situ, and not in a lab."))https://siliconangle.com/2026/01/16/openai-start-testing-chatgpt-ads-across-free-go-tiers/#fetched/018.txt L1
    sources/llms-full.txt#L69-70
    SiliconANGLE, January 16 2026, reports OpenAI will show ads to free and ChatGPT Go users; PARTIAL: the article's only pledge is 'not to share ChatGPT conversations with advertisers', a privacy promise, not an answers-not-influenced promise like Perplexity's.
    C-671So the checks are sealed before the run, held out from the agent, and run where the result is used, with the human's own act in the record.onedroid.ai/llms-full.txt (sources/llms-full.txt L79-80 (PAGE /; under "Why we test in situ, and not in a lab."))repo/internal/certificate/verify.go#L503
    repo/internal/role/role.go#L52
    repo/internal/role/role.go#L56
    The runner verifies seal and reveal anchors on chain and withholds scenarios and expected values from the product hat, so sealed-before and held-out are implemented for Argus Code; PARTIAL: 'run where the result is used' and 'the human's own act in the record' are not shown by any artefact.
    C-673Argus Code · the shape that exists todayonedroid.ai/llms-full.txt (sources/llms-full.txt L120 (PAGE /; under "Same modernization, three setups: what the tester checks decides whether it helps"))sources/github-repo.json#L1 private=false
    https://argus-dev.onedroid.ai#200
    sources/llms-full.txt#L120
    The public runner repository and the evaluation control plane answering 200 show an Argus Code shape exists today.
    C-676OneDroid Argus today tests AI-written code: a separate tester writes the checks, the coding agent never sees them, and they run against the deployed system.onedroid.ai/llms-full.txt (sources/llms-full.txt L3014-3015 (PAGE /argus-agents; under "Test the agents where they run,"))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L15-16 (under "Test the agents where they run,"))
    repo/internal/role/role.go#L52
    repo/internal/role/role.go#L56
    sources/github-repo.json#L1 description
    The runner redacts expected values and omits scenarios for the builder's hat, and the public repo describes it as executing sealed checks against your system; PARTIAL: that a separate tester writes them is a working practice, not an artefact.
    C-679Nothing on this page is available yet except what the Argus page already offers: the open-source runner and the hosted control plane. The two products below are in development.onedroid.ai/llms-full.txt (sources/llms-full.txt L3019-3020 (PAGE /argus-agents; under "Test the agents where they run,"))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L20-21 (under "Test the agents where they run,"))
    sources/github-repo.json#L1 private=false
    https://argus-dev.onedroid.ai#200
    sources/argus-agents.md#L20-21
    The two things said to be available exist (the public runner repository and the hosted control plane answering 200); PARTIAL: that nothing else on the page is available is an absence no artefact can show.
    C-680We publish what exists and what does not, and the claims page checks usonedroid.ai/llms-full.txt (sources/llms-full.txt L3021 (PAGE /argus-agents; under "Test the agents where they run,"))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L22 (under "Test the agents where they run,"))
    https://onedroid.ai/evidence/onedroid.ai#200
    sources/argus-agents.md#L60
    sources/argus-agents.md#L85
    The claims page answers 200 and the /argus-agents page carries explicit 'What exists today ... In development' status lines.
    C-681Every number OneDroid publishes is listed with its evidence, and a verdict is never a score: passed, failed, errored or not applicable, per check, with the evidence attached.onedroid.ai/llms-full.txt (sources/llms-full.txt L3038-3040 (PAGE /argus-agents; under "A certified agent can still be catastrophic once it is strung into a workflow."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L39-41 (under "A certified agent can still be catastrophic once it is strung into a workflow."))
    repo/internal/report/report.go#L404
    repo/internal/report/report.go#L577
    repo/internal/report/report.go#L123-132
    The report code has distinct passed, failed, error and not_applicable statuses per scenario, and the evidence page exists; PARTIAL: 'every number OneDroid publishes is listed' is a completeness claim not checkable here.
    C-688What exists today: the open-source Argus runner (MIT) and a cost-per-commit tool we use ourselves. In development: the merged binary, the session verdict, the credential monitor and the open check set.onedroid.ai/llms-full.txt (sources/llms-full.txt L3059-3061 (PAGE /argus-agents; under "One binary, on your machine, nothing leaves it."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L60-62 (under "One binary, on your machine, nothing leaves it."))
    repo/LICENSE#L1
    sources/github-repo.json#L1 license=mit
    sources/argus-agents.md#L60-62
    The open-source Argus runner under MIT exists; PARTIAL: the cost-per-commit tool is not public (repo-grep not-found), and the in-development list is a status statement.
    C-695What exists today: the Synapse gateway with its audit log, a recorder for Claude Code sessions in production on our own hubs, the Argus control plane and runner, and the ledger library. In development: the central store, the dashboards, the fleet alerts, the workflow scenarios and the replay.onedroid.ai/llms-full.txt (sources/llms-full.txt L3084-3087 (PAGE /argus-agents; under "Every agent recorded, every workflow tested, nothing leaves your estate."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L85-88 (under "Every agent recorded, every workflow tested, nothing leaves your estate."))
    https://synapse.onedroid.ai#200
    https://argus-dev.onedroid.ai#200
    sources/github-repo.json#L1 private=false
    The Synapse gateway, the Argus control plane and the open runner all exist; PARTIAL: the recorder for Claude Code sessions in production and the ledger library are not public, and the in-development list is a status statement.
    C-699Cells (one check judged on one system; a fixed check is judged on both systems, so a set has more cells than checks) | 21 | 32 |onedroid.ai/llms-full.txt (sources/llms-full.txt L3266 (PAGE /case-study-otrs-znuny; under "The result"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L89 (under "The result"))
    sources/case-study-otrs-znuny.md#L89
    sources/llms-full.txt#L3264-3274
    The result table's cell rows add up (identical 16 + differ 4 + noise 1 = 21; 23 + 9 + 0 = 32) and both exceed the check counts 16 and 26; PARTIAL: the number of fixed checks per set is not listed and the run reports are not published.
    C-728nothing of it is available yet, and the page says what exists and what does notonedroid.ai/llms.txt (sources/llms.txt L20 (under "OneDroid"))https://onedroid.ai/argus-agents.md#200 (claim check)
    sources/argus-agents.md#L20-21
    sources/argus-agents.md#L60-62
    /argus-agents does carry explicit 'What exists today' and 'In development' lines for both shapes; PARTIAL: 'nothing of it is available yet' is an absence restated, not shown.

    No evidence published (329)

    IdClaim, as madeMade atWhat would close it
    C-001Claude Code built a feature for $1.97 a run testing its own workonedroid.ai/llms-full.txt (sources/llms-full.txt L295 (PAGE /; under "What does independent testing cost?"))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L2750-2751 (PAGE /argus; under "Independent testing, at the same order of cost as the coding."))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L6154-6155 (PAGE /changelog; under "The site now leads with what testing costs"))
    The $1.97 mean appears only in a results table of means; the per-run tokens and costs are in a private bundle known only by its sha256 (L115), so no raw data is published.
    C-002With OneDroid Argus as the independent tester, the coding agent cost $1.01 a runonedroid.ai/llms-full.txt (sources/llms-full.txt L2751 (PAGE /argus; under "Independent testing, at the same order of cost as the coding."))
    onedroid.ai/argus.md (sources/argus.md L15 (under "Independent testing, at the same order of cost as the coding."))
    $1.01 is stated as a mean in the case-study table (1-1.01/1.97 = 0.487, consistent with -49%), but the per-run records behind it are private (sha256 only, L115).
    C-003the tester one session of about $5.30, which wrote both the build checks and the 13 hidden judge checks for the featureonedroid.ai/llms-full.txt (sources/llms-full.txt L295 (PAGE /; under "What does independent testing cost?"))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L2752 (PAGE /argus; under "Independent testing, at the same order of cost as the coding."))
    onedroid.ai/llms.txt (sources/llms.txt L41-42 (under "OneDroid"))
    The ~$5.30 tester session and the 13 hidden checks are stated on the page only; the judge checks and cost records are in the private bundle (sha256 only), so nothing replayable is published.
    C-004The tester is paid once per feature, and its checks stay as regression checksonedroid.ai/llms-full.txt (sources/llms-full.txt L295 (PAGE /; under "What does independent testing cost?"))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L2753 (PAGE /argus; under "Independent testing, at the same order of cost as the coding."))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L3660 (PAGE /case-study-claude-code-alone-vs-tester; under "What the numbers say"))
    Process claim stated on the home, Argus and case-study pages (the 37 older regression checks at L57 are also only stated); no published run record shows checks reused as regression checks.
    C-005Independent testing now costs the same order as the AI coding.onedroid.ai/llms-full.txt (sources/llms-full.txt L140 (PAGE /; under "50,000 lines overnight. How do you know what it did?"))A comparison resting on the one private case-study bundle ($2.78 vs $1.97); no public cost data backs the 'same order' assertion.
    C-006The coding agent cost 49% less with the testeronedroid.ai/llms.txt (sources/llms.txt L62 (under "Docs"))49% follows from the stated means ($1.01 vs $1.97, 1-1.01/1.97 = 0.487), but the means themselves rest on a private bundle (sha256 only), so the percentage is restated.
    C-007counting the tester's session, the whole cost 41% moreonedroid.ai/llms.txt (sources/llms.txt L62 (under "Docs"))41% is consistent with the stated means ($2.78/$1.97 = 1.411; $1.01 + $1.77 = $2.78), but no raw per-run costs are published (private bundle L115).
    C-008Both setups passed 13 of 13 hidden checks.onedroid.ai/llms.txt (sources/llms.txt L62 (under "Docs"))13 of 13 for both setups is stated in the case study, but the 13 judge checks and the per-run judge outputs are only in the private bundle (sha256 at L115).
    C-0093 judged runs per setup, in random order, on 16 October 2026onedroid.ai/llms-full.txt (sources/llms-full.txt L3626 (PAGE /case-study-claude-code-alone-vs-tester; under "The setup"))
    onedroid.ai/case-study-claude-code-alone-vs-tester.md (sources/case-study-claude-code-alone-vs-tester.md L63 (under "The setup"))
    The run count, random order and date are stated in the setup table only; no dated public run log or anchor for these runs is published.
    C-015It was not OneDroid Argus: our product's route could not reach the experiment's local setup.onedroid.ai/llms-full.txt (sources/llms-full.txt L3752 (PAGE /case-study-modernization-three-setups; under "The setup"))
    onedroid.ai/case-study-modernization-three-setups.md (sources/case-study-modernization-three-setups.md L61 (under "The setup"))
    A statement about the experiment's own tooling; no published configuration or record shows which tester ran or why Argus could not reach the setup.
    C-016A tester checking against the manual made the port worse (81 of 109), because the agent "fixed" every place where the original and its manual differ.onedroid.ai/llms-full.txt (sources/llms-full.txt L3712 (PAGE /case-study-modernization-three-setups; under "Same modernization, three setups: what the tester checks decides whether it helps"))
    onedroid.ai/case-study-modernization-three-setups.md (sources/case-study-modernization-three-setups.md L21 (under "Same modernization, three setups: what the tester checks decides whether it helps"))
    81 of 109 and the stated cause (B 'fixing' manual differences, 22 to 25 checks lost, L87) are only on the page; the ports and judge outputs are private.
    C-029A connection the architecture forbids was open. The cause is our cluster: it does not enforce Kubernetes network policies, so the demo's own policies had no effect there.onedroid.ai/llms-full.txt (sources/llms-full.txt L3410 (PAGE /case-study-finos-calm; under "In short"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L37 (under "In short"))
    The open forbidden connection and its traced cause (no network-policy enforcement on the cluster) are stated only; the manifests and network policies are in the private bundle.
    C-033The same sealed set of 26 checks, run three times against OTRS Community Edition 6.0.30 and three times against Znuny 7.3.7: six differences, what stayed the same, and the limits of the test.onedroid.ai/llms-full.txt (sources/llms-full.txt L3183 (PAGE /case-study-otrs-znuny))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L4 (under "Six differences between two releases of a ticketing system — OneDroid Argus case study"))
    The 26-check set and the two release versions are stated only; no sealed-set anchor, reveal or run record is published for this case study.
    C-034OneDroid Argus reported 23 cells identical, 9 that differ, none noisy and none not measured.onedroid.ai/llms-full.txt (sources/llms-full.txt L3210 (PAGE /case-study-otrs-znuny; under "In short"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L33 (under "In short"))
    23 identical + 9 differ = 32 cells matches the page's own table (L89-95), but no raw comparison record or anchor is published.
    C-035In one of them the older release returns an emoji as a replacement character, where the newer one returns it unchanged.onedroid.ai/llms-full.txt (sources/llms-full.txt L3212 (PAGE /case-study-otrs-znuny; under "In short"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L35 (under "In short"))
    The U+FFFD emoji difference (0 of 3 on OTRS, 3 of 3 on Znuny) is stated in the table only; no published run output shows it.
    C-036The sign-in page under five users at once: 95th percentile 49 ms on OTRS and 54 ms on Znuny in the first set, 60 ms and 59 ms in the second.onedroid.ai/llms-full.txt (sources/llms-full.txt L3316 (PAGE /case-study-otrs-znuny; under "What was the same"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L139 (under "What was the same"))
    The p95 sign-in figures are stated with no published load record.
    C-037The six runs of the first set took about 8 minutes.onedroid.ai/llms-full.txt (sources/llms-full.txt L3282 (PAGE /case-study-otrs-znuny; under "The result"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L105 (under "The result"))
    The ~8 minute duration is stated with no timestamps or run log published.
    C-038Three of the six differences were not found by a keyword search of the change log.onedroid.ai/llms-full.txt (sources/llms-full.txt L3368 (PAGE /case-study-otrs-znuny; under "What we take from it"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L191 (under "What we take from it"))
    The keyword-search result names the three fields (L123) but publishes no search terms or output, so it is restated.
    C-039it ran end to end for the first time on 4 October 2026, on three systemsonedroid.ai/llms-full.txt (sources/llms-full.txt L2905-2906 (PAGE /argus; under "One sealed set, two systems."))
    onedroid.ai/argus.md (sources/argus.md L169-170 (under "One sealed set, two systems."))
    The 4 October first end-to-end compare run on three systems is stated only; no dated run record or anchor for it is published.
    C-040An engineer on our team and his coding agent chose the two releases, wrote the checks and ran them; OneDroid Argus is our product.onedroid.ai/llms-full.txt (sources/llms-full.txt L3208 (PAGE /case-study-otrs-znuny; under "In short"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L31 (under "In short"))
    A statement of who did the work; nothing beyond the page backs it.
    C-042It does not judge the diff.onedroid.ai/llms-full.txt (sources/llms-full.txt L2774 (PAGE /argus; under "Yes, and it does more than a usual end-to-end suite."))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L2958 (PAGE /argus; under "Not a code reviewer"))
    onedroid.ai/argus.md (sources/argus.md L38 (under "Yes, and it does more than a usual end-to-end suite."))
    A negative design claim stated on the Argus page; the runner exposes no diff input, but absence is not an artefact, so it is restated.
    C-043A separate tester, your QA engineer or an agent but never the builder, writes each check from the ticket and the acceptance criteria, before the buildonedroid.ai/llms-full.txt (sources/llms-full.txt L2779 (PAGE /argus; under "Plain scenarios, not test scripts."))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L5431-5433 (PAGE /kit; under "Testing: OneDroid Argus"))
    onedroid.ai/argus.md (sources/argus.md L43 (under "Plain scenarios, not test scripts."))
    Who writes the checks and when is a process claim; the repo's separate author and product hats support role separation but not that checks are written from the ticket before the build.
    C-047The hash proves priority. Keeping them unseen is a separate mechanismonedroid.ai/llms-full.txt (sources/llms-full.txt L2820-2821 (PAGE /argus; under "Sealed before the build."))
    onedroid.ai/argus.md (sources/argus.md L84-85 (under "Sealed before the build."))
    A design statement about what a hash does and does not prove; nothing beyond the statement backs the separate 'unseen' mechanism as such.
    C-063No load ramp sits under a certified verdict or on a schedule.onedroid.ai/llms-full.txt (sources/llms-full.txt L2881 (PAGE /argus; under "Under load."))
    onedroid.ai/argus.md (sources/argus.md L145 (under "Under load."))
    A status about practice; the runner code allows ramps in certifying modes (only logging differs, L308-311), so nothing published shows that no ramp sits under a certified verdict or schedule.
    C-065the reviewed scenario set goes on a monitor schedule: every 15 minutes at the shortest, with no one driving itonedroid.ai/llms-full.txt (sources/llms-full.txt L2888-2889 (PAGE /argus; under "Continuously."))
    onedroid.ai/argus.md (sources/argus.md L152-153 (under "Continuously."))
    The monitor schedule is a hosted control-plane feature; the open repo has no 15-minute schedule minimum (grep: only token lifetimes), so it is stated only.
    C-066The web app answers one question first, “Is it OK?”, from the scheduled runs only: a failing run someone started by hand never turns it red, and a system no monitor has measured reads “not measured”, never green.onedroid.ai/llms-full.txt (sources/llms-full.txt L2889-2891 (PAGE /argus; under "Continuously."))
    onedroid.ai/argus.md (sources/argus.md L153-155 (under "Continuously."))
    'Is it OK?' is not in the open repo (repo-grep not-found); the web app is closed and behind sign-in, so the claim is stated only.
    C-067What the system cannot fix by itself is listed under “Needs a person”, with who acts next.onedroid.ai/llms-full.txt (sources/llms-full.txt L2891-2892 (PAGE /argus; under "Continuously."))
    onedroid.ai/argus.md (sources/argus.md L155-156 (under "Continuously."))
    'Needs a person' is not in the open repo (repo-grep not-found); it is a closed web-app feature stated only.
    C-078Release binaries are built for macOS and Linux. On Windows, use WSL 2.onedroid.ai/llms-full.txt (sources/llms-full.txt L2921 (PAGE /argus; under "CLI"))
    onedroid.ai/argus.md (sources/argus.md L185 (under "CLI"))
    The public runner repo has no releases (empty list) and the quickstart says the release repository is private, so the macOS/Linux binaries cannot be checked.
    C-082The OneDroid Argus page now describes release 0.3.70.onedroid.ai/llms-full.txt (sources/llms-full.txt L6244-6245 (PAGE /changelog; under "OneDroid Argus 0.3.70: a check can read the database and the logs, and make the system fail"))
    onedroid.ai/changelog.md (sources/changelog.md L135-136 (under "OneDroid Argus 0.3.70: a check can read the database and the logs, and make the system fail"))
    The Argus page has no version number (grep 0.3.x: none) and no public release tag exists; repo comments name executor 0.3.70 for the features described, so the version claim is restated.
    C-083From v0.3.37 to v0.3.51 every load duration was ignored, so load numbers taken in that range are not measured: re-run them on v0.3.52.docs.onedroid.ai/argus (sources/argus-docs.html L321-322 (under "What changed recently"))
    docs.onedroid.ai/argus.md (sources/argus-docs.md L241-242 (under "What changed recently"))
    The bug range is stated in the docs changelog only; the public repo has a single squashed commit and no fix record for load durations.
    C-085install the GitHub app, link a repository, and every successful deployment runs your checks and puts the verdict on the commit.docs.onedroid.ai/argus (sources/argus-docs.html L363-364 (under "Where to go next"))
    docs.onedroid.ai/argus.md (sources/argus-docs.md L276-277 (under "Where to go next"))
    The GitHub app integration is a hosted control-plane feature; no app, workflow or commit-status code is in the open repo (grep: no match).
    C-093the control plane that seals the set and adjudicates the verdict is hosted and closedonedroid.ai/llms-full.txt (sources/llms-full.txt L269-270 (PAGE /; under "Open where it has to be."))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L5941-5942 (PAGE /pricing; under "The OneDroid Argus control plane"))
    onedroid.ai/pricing.md (sources/pricing.md L56-57 (under "The OneDroid Argus control plane"))
    The repo README confirms the control plane is hosted and not in the open repo, but that it is closed everywhere cannot be checked from public surfaces (the org repo listing answered 403).
    C-096The enrollment token lives 900 secondsraw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L70 (under "Quickstart"))The 900-second lifetime is set by the closed control plane that mints the token; the open repo only repeats it in a comment and a test fixture, so nothing independent shows the real TTL (a minted token's expires_in would be needed).
    C-097Anyone can sign up for the evaluation control plane at https://argus-dev.onedroid.ai (a test environment: no promise that data is kept)onedroid.ai/llms.txt (sources/llms.txt L46-47 (under "OneDroid"))The evaluation control plane answers 200 and its bundle wires a Clerk sign-up redirect, but open self-service sign-up is not observable without creating an account (forbidden) and the bundle carries no 'no promise that data is kept' text, so the claim rests on the pages.
    C-098an environment of your own is by requestonedroid.ai/llms-full.txt (sources/llms-full.txt L5294 (PAGE /qa-teams; under "Bring one feature."))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L6024 (PAGE /pricing; under "What does OneDroid Argus cost?"))
    onedroid.ai/llms.txt (sources/llms.txt L47 (under "OneDroid"))
    A process statement repeated on /qa-teams, /pricing and llms.txt; no request form, contract or other artefact shows how a dedicated environment is provisioned.
    C-099The hosted control plane that seals the withheld scenarios and adjudicates the verdict is the priced part.onedroid.ai/llms-full.txt (sources/llms-full.txt L6024 (PAGE /pricing; under "What does OneDroid Argus cost?"))
    onedroid.ai/pricing.md (sources/pricing.md L139 (under "What does OneDroid Argus cost?"))
    The runner README confirms the control plane is closed and outside the repo, but no price list or billing artefact shows it is the priced part ('no public price list for the enterprise tier', pricing.md L135).
    C-100The Argus release repository is private. Ask your Argus operator for read access to itdocs.onedroid.ai/argus-quickstart (sources/argus-quickstart.html L76-77 (under "1. Get the argus CLI"))
    docs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L27-28 (under "1. Get the `argus` CLI"))
    The page withholds the release repository's owner/name ('your operator gives you its owner/name'), so no public lookup can confirm it is private; only the statement exists.
    C-103A governed MCP gateway. Your AI agents connect to one URL instead of holding their own credentials for each service, and every tool call is authenticated, checked against policy, and logged while the hub's activity logging is on (fail-open: a logging failure never blocks a call), to an audit log you own.onedroid.ai/llms-full.txt (sources/llms-full.txt L4168 (PAGE /synapse; under "What is OneDroid Synapse?"))The live endpoint shows an authenticated MCP URL, but the parts that make it governed (policy check per call, logging to an audit log the customer owns, fail-open logging) live in closed Synapse code with no public artefact.
    C-104Per-tool access control, not just per-integrationonedroid.ai/llms.txt (sources/llms.txt L109 (under "Key capabilities"))Per-tool access control is a Synapse server feature behind sign-in; no public code, schema or endpoint shows it (a public policy schema or docs page with a per-tool rule would be needed).
    C-105Agents never hold long-lived upstream credentials. The gateway holds them and presents them per callonedroid.ai/llms-full.txt (sources/llms-full.txt L4244-4245 (PAGE /what-is-a-governed-mcp-gateway; under "Credential custody"))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L4707-4708 (PAGE /security; under "Credential custody"))
    onedroid.ai/security.md (sources/security.md L98-99 (under "Credential custody"))
    Credential custody is stated on /what-is-a-governed-mcp-gateway and /security; the gateway is closed and no public artefact shows credentials held server-side and presented per call.
    C-106An append-only event log of every tool invocation while the hub's activity logging is on — caller, tool, arguments hash, latency, result class. Yours to query and export.onedroid.ai/llms-full.txt (sources/llms-full.txt L4017-4018 (PAGE /synapse; under "Audit log"))The append-only event log and its fields are described only; repo grep for append-only finds nothing (the open repo is Argus, not Synapse) and the log needs an account to view.
    C-107The log and its CSV export are included in the free tier and will not be put behind a paywallonedroid.ai/llms-full.txt (sources/llms-full.txt L4712-4713 (PAGE /security; under "Append-only audit log"))
    onedroid.ai/security.md (sources/security.md L103-104 (under "Append-only audit log"))
    A forward-looking pricing commitment repeated on /security and /pricing; nothing beyond the statement can back a promise about the future.
    C-108Synapse and Engram are free on the hosted shape, whatever the size of your teamonedroid.ai/llms-full.txt (sources/llms-full.txt L5891 (PAGE /pricing))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L5901-5902 (PAGE /pricing; under "Free to use."))
    onedroid.ai/llms.txt (sources/llms.txt L92 (under "Product"))
    The pricing page states it and the accounts instance has Clerk billing switched off, which is consistent, but no artefact shows the hosted service is actually unpriced at every team size.
    C-109The boundary is deployment shape, not seat count.onedroid.ai/llms-full.txt (sources/llms-full.txt L6016 (PAGE /pricing; under "Where does the free tier end?"))
    onedroid.ai/pricing.md (sources/pricing.md L131 (under "Where does the free tier end?"))
    The tier boundary is the publisher's own pricing statement, repeated in the tier table and FAQ; no billing or contract artefact independent of the page shows enterprise begins at deployment shape rather than seats.
    C-110If usage limits ever arrive, they'll be published here firstonedroid.ai/llms-full.txt (sources/llms-full.txt L6002-6003 (PAGE /pricing; under "Enterprise"))
    onedroid.ai/pricing.md (sources/pricing.md L117-118 (under "Enterprise"))
    A promise about future publication; nothing can evidence it beyond the statement.
    C-112Dedicated deployment in your cloud or on-premises, under license.onedroid.ai/llms-full.txt (sources/llms-full.txt L5790 (PAGE /enterprise; under "Your infrastructure"))Dedicated deployment under licence is stated on /enterprise; no public licence text, deployment artefact or customer reference backs it (repo grep for on-premises finds nothing).
    C-117It is included with a Synapse account and runs on its own.onedroid.ai/llms-full.txt (sources/llms-full.txt L4474 (PAGE /quickstart; under "Give your agents memory that outlives the session."))Engram's inclusion with a Synapse account is stated on /quickstart; no public artefact shows the account bundling.
    C-118Results are fused into a single ranking, and the vector-versus-keyword weighting is a parameter you control per query — not a black box.onedroid.ai/llms-full.txt (sources/llms-full.txt L4888-4889 (PAGE /engram; under "Graph"))Fused ranking and a per-query vector/keyword weight are described on /engram; the Engram tool schema is behind a token and no public code or docs page shows the parameter.
    C-119Objects carry revision history. Diff any two versions, restore a previous one in one call.onedroid.ai/llms-full.txt (sources/llms-full.txt L4520-4521 (PAGE /quickstart; under "Every write is a commit"))Revision history, diff and one-call restore are described on /quickstart; no public tool schema or endpoint shows them without an Engram token.
    C-120Point Engram at your own Postgres and documents, chunks, embeddings and history are rows there, readable with SQL. It needs the pgvector extension, which Engram enables at registration; Postgres 16 is what we test against.onedroid.ai/llms-full.txt (sources/llms-full.txt L4514-4516 (PAGE /quickstart; under "Your own database"))Bring-your-own-Postgres with pgvector is described only; no public schema, migration or code shows the tables or the extension being enabled at registration.
    C-121Export is a SQL queryonedroid.ai/llms-full.txt (sources/llms-full.txt L4060 (PAGE /synapse; under "No lock-in"))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L5001 (PAGE /financial-services; under "Operations"))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L5052 (PAGE /financial-services; under ""What happens when we leave?""))
    Export by SQL is repeated on /synapse and /financial-services; no published schema or query backs it.
    C-122Curated toolsets with the context to use them safely — tools, skills, and knowledge shipped together, versioned, and scoped per hub.onedroid.ai/llms-full.txt (sources/llms-full.txt L4007-4008 (PAGE /synapse; under "Packs"))Packs are described on /synapse; no public pack manifest or catalogue endpoint shows versioned, per-hub toolsets.
    C-123Governance, as switches. PHI detection, credential encryption, an append-only audit trail, long-term compliance logging, rate limiting, IP allowlisting. The padlocked rows are forced on by a platform-wide override and cannot be turned off for an individual hubonedroid.ai/llms-full.txt (sources/llms-full.txt L4147-4150 (PAGE /synapse; under "Not a mockup."))The governance switches and padlocked platform override are described (with a screenshot reference) but the settings sit behind sign-in; repo and source greps for append-only, long-term compliance and platform-wide find nothing.
    C-124Credentials bound per user, per hub — encrypted at rest with AES-256-GCM, individually revocable, never sitting in a connection string.onedroid.ai/llms-full.txt (sources/llms-full.txt L5021-5022 (PAGE /financial-services; under ""Whose credentials is the agent using?""))
    onedroid.ai/financial-services.md (sources/financial-services.md L107-108 (under ""Whose credentials is the agent using?""))
    AES-256-GCM per-user, per-hub credential binding is stated; no public code or security document shows the encryption scheme.
    C-125A hash chain, a Merkle tree, canonical serialisation, a transactional outbox so writes can't be lost, a reconciler, a circuit breaker — and an independent verifier, because a system attesting to itself proves nothing.onedroid.ai/llms-full.txt (sources/llms-full.txt L5033-5036 (PAGE /financial-services; under ""How do I know the log wasn't edited?""))
    onedroid.ai/financial-services.md (sources/financial-services.md L119-122 (under ""How do I know the log wasn't edited?""))
    The hash chain, Merkle tree, outbox, reconciler, circuit breaker and independent verifier for the Synapse log are listed only; the Synapse audit code is not public (the open Argus repo's Merkle reveal code is a different component).
    C-126Detection in the request path, returning a redaction reference for audit — not a nightly batch job that finds it after the fact. And tuned rather than naive: ours excludes date patterns from free-text lanes, because specification documents are full of dates.onedroid.ai/llms-full.txt (sources/llms-full.txt L5040-5043 (PAGE /financial-services; under ""What stops client data reaching the model?""))
    onedroid.ai/financial-services.md (sources/financial-services.md L126-129 (under ""What stops client data reaching the model?""))
    In-path PHI detection with a redaction reference and the date-pattern exclusion are described; no public code or test shows them (repo grep for free-text lanes finds nothing).
    C-127A free personal hub and a regulated enterprise deployment use the same mechanism.onedroid.ai/llms-full.txt (sources/llms-full.txt L4065-4066 (PAGE /synapse; under "Any scale"))A checkability-C architectural statement; nothing public shows that a personal hub and an enterprise deployment run the same mechanism.
    C-128Synapse's audit entry, Engram's versioned fact and OneDroid Argus's verdict are the same object — a promise plus the evidence for itonedroid.ai/llms-full.txt (sources/llms-full.txt L243-244 (PAGE /; under "When someone has to answer for it."))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L5855-5857 (PAGE /enterprise; under "One primitive: the evidence record"))
    A checkability-C framing that three products share one evidence object; no shared schema or format document is published.
    C-129Hosted: tool-call metadata and encrypted credentials. Hosted with your own database: the same, minus your OneDroid Engram content. Self-hosted: nothing at runtime.onedroid.ai/llms-full.txt (sources/llms-full.txt L4184 (PAGE /synapse; under "What can OneDroid see?"))What OneDroid can see per deployment shape is stated on /synapse; no data-flow artefact or code shows it.
    C-133In two quarters it went from "can we even use AI here" to a running, audited platform.onedroid.ai/llms-full.txt (sources/llms-full.txt L5105-5106 (PAGE /financial-services; under "Proven where the audits are unforgiving."))
    onedroid.ai/financial-services.md (sources/financial-services.md L191-192 (under "Proven where the audits are unforgiving."))
    An unnamed-customer outcome with no case study, timeline or reference behind it.
    C-134We do not hold SOC 2, ISO 27001, or HIPAA attestation today.onedroid.ai/llms-full.txt (sources/llms-full.txt L4761-4762 (PAGE /security; under "What is not yet true"))
    onedroid.ai/security.md (sources/security.md L152-153 (under "What is not yet true"))
    A self-declared absence of SOC 2, ISO 27001 and HIPAA attestation; no page shows an attestation report or badge, which is consistent, but an absence has no artefact of its own.
    C-135Credentials are encrypted at rest. Transport is TLS. The event log is append-only and exportable.onedroid.ai/llms-full.txt (sources/llms-full.txt L4755-4756 (PAGE /security; under "What is true"))
    onedroid.ai/security.md (sources/security.md L146-147 (under "What is true"))
    TLS is observable on every host fetched over https, but encryption at rest and an append-only exportable log have no public artefact, so the three-part statement rests on the page.
    C-136The OneDroid application has completed Google's OAuth application verification, including its branding and app-purpose review, for the Google Workspace connections it offers.onedroid.ai/llms-full.txt (sources/llms-full.txt L4729-4731 (PAGE /security; under "Standards, not a bespoke scheme."))
    onedroid.ai/security.md (sources/security.md L120-122 (under "Standards, not a bespoke scheme."))
    Google OAuth verification is stated; no public artefact on the publisher's surfaces (a consent-screen capture or verification notice) shows it, and the Google side was not fetched.
    C-139We reply within one business day, agree a disclosure timeline with you, and credit you unless you would rather we didn't.onedroid.ai/llms-full.txt (sources/llms-full.txt L4778-4779 (PAGE /security; under "Tell us privately, first."))
    onedroid.ai/security.md (sources/security.md L169-170 (under "Tell us privately, first."))
    A disclosure-handling commitment; no published security.txt policy history, hall of credit or response record backs it.
    C-140We reply within one business day.onedroid.ai/llms-full.txt (sources/llms-full.txt L318 (PAGE /; under "Find out what your agent actually built."))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L2998 (PAGE /argus; under "Find out what your agent actually built."))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L3108 (PAGE /argus-agents; under "Tell us which workflow you would test first."))
    The one-business-day reply is repeated on many pages; no response-time record exists to back it.
    C-141There is no sales queue. Email reaches Michal Bacia, who writes the codeonedroid.ai/llms-full.txt (sources/llms-full.txt L6085 (PAGE /about; under "A small team, reachable directly."))The open repo has one committer identity ('MB'), which is consistent with one person writing the code, but nothing ties that identity to the contact address or shows there is no sales queue.
    C-142you get direct access to the people building it, and in exchange you are not buying from a company with a 24/7 support rota. If that matters for your deployment, tell us early and we will be honest about what we can commit to.onedroid.ai/llms-full.txt (sources/llms-full.txt L6089-6093 (PAGE /about; under "A small team, reachable directly."))A statement about team size and support arrangements; nothing public evidences it.
    C-143We do not sell your data. We do not use your data for advertising. We do not train AI models on your data.onedroid.ai/llms-full.txt (sources/llms-full.txt L6422-6423 (PAGE /privacy; under "The short version"))
    onedroid.ai/privacy.md (sources/privacy.md L31-32 (under "The short version"))
    Privacy-policy commitments about data use; no audit, contract or technical artefact can show the absence of selling, advertising or training.
    C-144Humans at OneDroid do not read the content that passes through the gateway or that you store in the Service — whether it comes from Google, another connected service, or OneDroid Engram — except with your explicit permission, where necessary for security or abuse investigation, or where required by law.onedroid.ai/llms-full.txt (sources/llms-full.txt L6472-6475 (PAGE /privacy; under "Access to your content"))
    onedroid.ai/privacy.md (sources/privacy.md L81-84 (under "Access to your content"))
    A human-access policy statement; no access-control artefact or audit is published.
    C-146Audit log entries on the free hosted tier are retained for 12 months and then deleted, and are deleted within 30 days of account deletiononedroid.ai/llms-full.txt (sources/llms-full.txt L6543-6544 (PAGE /privacy; under "Retention and deletion"))
    onedroid.ai/privacy.md (sources/privacy.md L152-153 (under "Retention and deletion"))
    Retention periods are stated in the privacy policy; no retention configuration or deletion record is public.
    C-148a human reads it and replies within one business day; the request itself is completed within the timelines applicable law requires.onedroid.ai/llms-full.txt (sources/llms-full.txt L6548 (PAGE /privacy; under "Retention and deletion"))
    onedroid.ai/privacy.md (sources/privacy.md L157 (under "Retention and deletion"))
    A response-time and handling commitment for data requests; no record backs it.
    C-149Google OAuth tokens are stored encrypted at rest. Google user content is not retained by the Service beyond serving your request, except where you explicitly store itonedroid.ai/llms-full.txt (sources/llms-full.txt L6494-6496 (PAGE /privacy; under "Google user data"))
    onedroid.ai/privacy.md (sources/privacy.md L103-105 (under "Google user data"))
    Encryption of Google OAuth tokens and non-retention of Google content are stated; the Synapse code is closed and no public artefact shows either.
    C-150The Service is not directed to children under 16, and we do not knowingly collect their data.onedroid.ai/llms-full.txt (sources/llms-full.txt L6570-6571 (PAGE /privacy; under "Children"))
    onedroid.ai/privacy.md (sources/privacy.md L179-180 (under "Children"))
    A policy statement about children's data; nothing beyond it can be shown (the Clerk sign-up config carries no age gate, which neither backs nor refutes it).
    C-152Providentia Worldwide is an engineering consultancy that works inside regulated enterprise estates — health data, financial tradingonedroid.ai/llms-full.txt (sources/llms-full.txt L6051-6053 (PAGE /about; under "OneDroid is built by Providentia Worldwide."))A company description on /about with no named engagements, case studies or registry link behind it.
    C-153Providentia Worldwide holds five granted United States patents on interlinked, N-dimensional blockchain structures and globally verifiable identifiers (US 10,853,354; 11,023,458; 11,604,787; 12,141,127; 12,493,609), with further applications pending in the US and Europe.onedroid.ai/llms-full.txt (sources/llms-full.txt L6062-6065 (PAGE /about; under "OneDroid is built by Providentia Worldwide."))Five US patent numbers are listed, but no patent document is linked or was fetched for this run, so the grant and assignee are not shown on the publisher's surfaces (the USPTO records would be needed).
    C-154The audit trail behind the OneDroid evidence record is designed on that structure.onedroid.ai/llms-full.txt (sources/llms-full.txt L6065-6066 (PAGE /about; under "OneDroid is built by Providentia Worldwide."))A design-lineage statement; no design document or code ties the evidence-record audit trail to the patented structure.
    C-159For years companies spent between 23% and 35% of their IT budget on quality assurance and testing (23% in 2013, 26% in 2014, 35% in 2015, 31% in 2016, 26% in 2017). When it was first measured at 18%, Capgemini's World Quality Report called that "an indication of insufficient attention on quality".onedroid.ai/llms-full.txt (sources/llms-full.txt L2797-2798 (PAGE /argus; under "Testing was the part teams could skip. Not any more."))
    onedroid.ai/argus.md (sources/argus.md L61-62 (under "Testing was the part teams could skip. Not any more."))
    Capgemini World Quality Report figures are cited by name with no link, and no copy was fetched for this run, so the series and the quoted phrase have no artefact on the publisher's surfaces.
    C-16080% have traced a production incident directly back to AI-generated code, and 92% aren't confident their current safeguards would catch an AI-driven failure before it reaches users. Sauce Labs, “The Enterprise AI Code Verification Crisis”, n = 400 US executives and engineering leaders, 22 July 2026 — a vendor-commissioned survey.onedroid.ai/llms-full.txt (sources/llms-full.txt L144-148 (PAGE /; under "You already had the incident."))The Sauce Labs survey figures carry a title, n and date but no link and no fetched copy (the page cites an internal 'AS-9 §2'), so nothing replayable backs them.
    C-1613.1% of developers highly trust AI output; 19.6% highly distrust it; 66% hit answers that are “almost right, but not quite”. Stack Overflow Developer Survey 2025, n = 33,244.onedroid.ai/llms-full.txt (sources/llms-full.txt L151-153 (PAGE /; under "You already had the incident."))Stack Overflow 2025 survey figures are attributed with n = 33,244 but not linked and not fetched for this run.
    C-162Syntax pass rates for AI-generated code climbed from about 50% to 95%. Security pass rates stayed flat, between 45% and 55%. Veracode, Spring 2026 GenAI Code Security update, 150+ models across 80 coding tasks, 24 March 2026.onedroid.ai/llms-full.txt (sources/llms-full.txt L155-158 (PAGE /; under "You already had the incident."))Veracode Spring 2026 figures are attributed with model and task counts and a date, but not linked and not fetched for this run.
    C-163Claude Code Review's check “always completes with a neutral conclusion so it never blocks merging”; Copilot's reviews “do not count toward required approvals”.onedroid.ai/llms-full.txt (sources/llms-full.txt L161-163 (PAGE /; under "You already had the incident."))Two vendor-documentation quotes are given without links ('Vendor documentation, AS-9 §1 and §5'); no copy was fetched, so the quotes are not shown on any surface.
    C-164the largest vendor's hosted agents say of themselves that they are “not currently eligible for Zero Data Retention (ZDR) or HIPAA Business Associate Agreement (BAA) coverage.”onedroid.ai/llms-full.txt (sources/llms-full.txt L250-252 (PAGE /; under "When someone has to answer for it."))A quoted vendor statement attributed to 'the largest vendor' with no link or name; no copy was fetched.
    C-165EU AI Act Art. 12 (automatic recording of events over the lifetime of the system) and Art. 14 (human oversight) for high-risk AI products; DORA Art. 25 and Art. 30; HIPAA 164.312(b); FDA 21 CFR 11.10(e). Using a coding agent does not by itself make your system high-risk, and we will not tell you otherwise.onedroid.ai/llms-full.txt (sources/llms-full.txt L245-248 (PAGE /; under "When someone has to answer for it."))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L5862-5866 (PAGE /enterprise; under "One primitive: the evidence record"))
    Regulatory article citations are named but not linked, and no mapping document shows how the evidence record meets each article; the 'not high-risk by itself' line is a stance with nothing to evidence.
    C-166OneDroid (onedroid.ai) builds AI agent infrastructure. It is unrelated to Android and to the open-source Android projects also called "OneDroid".onedroid.ai/llms-full.txt (sources/llms-full.txt L9-10 (under "OneDroid — full site text for language models"))
    onedroid.ai/llms.txt (sources/llms.txt L8-9 (under "OneDroid"))
    A self-description and disambiguation; nothing on the surfaces can evidence unrelatedness beyond the statement.
    C-169Entries are added when something ships, not when it is planned.onedroid.ai/llms-full.txt (sources/llms-full.txt L6326 (PAGE /changelog; under "onedroid.ai launched"))
    onedroid.ai/changelog.md (sources/changelog.md L217 (under "onedroid.ai launched"))
    An editorial rule for the changelog; nothing beyond the statement shows entries were added only on shipping.
    C-170We publish limitations alongside capabilities and say what doesn't work yet. If a claim can't be verified, we don't make it.onedroid.ai/llms-full.txt (sources/llms-full.txt L4137-4138 (PAGE /synapse; under "Honesty"))A checkability-C honesty commitment; a limitations section exists on /security, but no artefact can show the general rule holds.
    C-172Every call on those pages was run against the live service.onedroid.ai/llms-full.txt (sources/llms-full.txt L4466 (PAGE /quickstart; under "The check"))The no-token responses observed match the quickstart pages, which is consistent, but no run log shows that every documented call was run against the live service.
    C-173Each takes about ten minutes, needs no credit card, and ends with a check you run yourselfonedroid.ai/llms-full.txt (sources/llms-full.txt L4389-4390 (PAGE /quickstart; under "Three things to try today."))Time, no-card and self-check are stated; the checks exist on the page but the duration and no-card sign-up cannot be shown without creating an account.
    C-174Your data is never used for advertising and never used to train AI models, and access can be revoked at any time in OneDroid Synapse or in your Google account settings.onedroid.ai/llms-full.txt (sources/llms-full.txt L3950-3952 (PAGE /synapse; under "What OneDroid Synapse does."))Data-use and revocation statements on /synapse; no artefact shows them (the revocation UI is behind sign-in).
    C-175logged while the hub's activity logging is on (fail-open: a logging failure never blocks a call)onedroid.ai/llms-full.txt (sources/llms-full.txt L228 (PAGE /; under "It brings its own run layer."))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L3909 (PAGE /synapse))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L3923 (PAGE /synapse; under "OneDroid Synapse"))
    The fail-open logging clause is repeated on twelve locations, but Synapse code is closed and repo/source greps for fail-open find nothing.
    C-177The flagship deployment is a US enterprise handling regulated clinical dataonedroid.ai/llms-full.txt (sources/llms-full.txt L5103 (PAGE /financial-services; under "Proven where the audits are unforgiving."))
    onedroid.ai/financial-services.md (sources/financial-services.md L189 (under "Proven where the audits are unforgiving."))
    An unnamed flagship customer; no case study, reference or named organisation backs it.
    C-178No named financial-services customer yet [...] The flagship reference is in regulated healthcareonedroid.ai/llms-full.txt (sources/llms-full.txt L5125-5127 (PAGE /financial-services; under "No named financial-services customer yet"))
    onedroid.ai/financial-services.md (sources/financial-services.md L211-213 (under "No named financial-services customer yet"))
    A self-declared absence of a named financial-services customer and an unnamed healthcare reference; nothing beyond the statement.
    C-194**A new web app** (v0.3.53): five places, **Overview**, **Capacity**, **Runs**, **Checks** and **Proof**, one run drawer, and **Needs a person** items you file with `author_file_attention`.docs.onedroid.ai/argus.md (sources/argus-docs.md L226-227 (under "What changed recently"))The web app is part of the closed hosted control plane; author_file_attention and the five places are not in the open repo and a signed-out GET shows only the app shell, so it needs a signed-in view or published UI build to check.
    C-198**Run** | whichever scenarios you ask for, once | authoring and iterating | | **Monitor** | your reviewed scenario set, on a schedule | watching a live system stay healthydocs.onedroid.ai/argus.md (sources/argus-docs.md L92-93 (under "The two ways to start a run"))The open skill names author_request_run and author_set_schedule on a sealed set, but the scheduler is in the closed control plane; a public schedule artefact would be needed.
    C-199Hand-started runs can say what they are for with `intent`, `expect` and `note`.docs.onedroid.ai/argus.md (sources/argus-docs.md L228 (under "What changed recently"))Run-level intent, expect and note fields belong to the closed control plane's request tool and were not found in the open repo or another public surface.
    C-211For another platform, change `linux_amd64` to `linux_arm64`, `darwin_amd64` (Intel Mac) or `darwin_arm64` (Apple silicon).docs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L43-44 (under "1. Get the `argus` CLI"))The install text points at a release repository that needs read access, the public argus-runner has no latest release and the clone has no build matrix, so the four platform archives could not be seen.
    C-216Access exists only for accounts you explicitly connect, is limited to the permissions you grant, and every action is logged while the hub's activity logging is on (fail-open: a logging failure never blocks a call), to an audit log you own. Disconnect an account at any time and OneDroid Synapse's access ends with it.onedroid.ai/llms-full.txt (sources/llms-full.txt L3943-3945 (PAGE /synapse; under "What OneDroid Synapse does."))Synapse's per-account access, fail-open logging and disconnect behaviour are stated only; the gateway code is not public and the open repo has no fail-open match, so a public audit-log spec or code would be needed.
    C-217Tokens are bound to one user and one hub, and inherit that user's current role in the hub.onedroid.ai/llms.txt (sources/llms.txt L122-123 (under "Connecting an agent"))Token binding to one user and one hub with inherited role is stated only; it would need public gateway code or an API reference that shows it.
    C-219Enabling a service on your hub and saving your login for it are two different actions, and only the second lets an agent call anything.onedroid.ai/llms-full.txt (sources/llms-full.txt L4438-4439 (PAGE /quickstart; under "The step that catches everyone"))The enable-versus-save-login distinction is stated in the quickstart only; it would need a signed-in walkthrough or public code to check.
    C-220It reaches one library and nothing else [...] Or reach it through your Synapse hub, where the Engram connection is already on and every call is audited with the rest of your tools.onedroid.ai/llms-full.txt (sources/llms-full.txt L4481-4491 (PAGE /quickstart; under "Get a library token"))Library-scoped Engram tokens and the hub's pre-wired, audited Engram connection are stated only; no public artefact shows the scoping.
    C-221Tool payloads pass through the gateway in order to be routed. [...] Self-hosted deployment removes us from the data path entirely.onedroid.ai/llms-full.txt (sources/llms-full.txt L4678-4757 (PAGE /security; under "Hosted"))
    onedroid.ai/security.md (sources/security.md L69-148 (under "Hosted"))
    The data-path description of hosted and self-hosted deployment is stated on /security only; it would need a published architecture or code.
    C-223Credentials held per user, per hub. Sensitive data detected in the request path rather than in a batch job afterwards; in the reference deployment some controls run in audit mode rather than blocking.onedroid.ai/llms-full.txt (sources/llms-full.txt L4991 (PAGE /financial-services; under "Security"))
    onedroid.ai/financial-services.md (sources/financial-services.md L77 (under "Security"))
    Per-user, per-hub credentials and in-path sensitive-data detection in audit mode are stated only; they would need gateway code or a published config.
    C-224Tool-level allow and deny policy, with the decision recorded at the moment it was made — not reconstructed from logs afterwardsonedroid.ai/llms-full.txt (sources/llms-full.txt L5027-5028 (PAGE /financial-services; under ""What can this agent reach, and who decided?""))
    onedroid.ai/financial-services.md (sources/financial-services.md L113-114 (under ""What can this agent reach, and who decided?""))
    Tool allow/deny with the decision recorded at call time is stated only; it would need a public policy schema or an audit-log sample.
    C-226we hash rather than store them in the audit log, and we work to keep content out of operational logs, though we cannot promise that no fragment ever appears in error telemetryonedroid.ai/llms-full.txt (sources/llms-full.txt L6449-6451 (PAGE /privacy; under "What we collect"))
    onedroid.ai/privacy.md (sources/privacy.md L58-60 (under "What we collect"))
    Hashing in the audit log and keeping content out of operational logs are stated only; they would need gateway code or a published log schema.
    C-227Our subprocessors operate in the United States and, for Microsoft Azure hosting, in the European Union (Poland Central); if you need documentation covering international transfers, or a data processing agreement for your team, email michal@onedroid.ai.onedroid.ai/llms-full.txt (sources/llms-full.txt L6554-6556 (PAGE /privacy; under "Where GDPR or CCPA applies"))
    onedroid.ai/privacy.md (sources/privacy.md L163-165 (under "Where GDPR or CCPA applies"))
    Subprocessor locations (US, Azure Poland Central) are stated only; a published subprocessor list or DPA would be needed.
    C-228If we ever discontinue the free hosted Service, we will give reasonable notice and a window to export your data. [...] the free Service is not a backup systemonedroid.ai/llms-full.txt (sources/llms-full.txt L6624-6652 (PAGE /terms; under "2. The Service"))A notice-and-export promise in the terms; there is nothing to check until the Service is discontinued.
    C-229Effective date: 19 July 2026 · Last updated: 14 October 2026onedroid.ai/llms-full.txt (sources/llms-full.txt L6405 (PAGE /privacy; under "Privacy Policy"))
    onedroid.ai/privacy.md (sources/privacy.md L14 (under "Privacy Policy"))
    The changelog dates the privacy policy's publication 2026-07-19, matching the effective date; 'Last updated 14 October 2026' would need a public revision history of the policy.
    C-230Enterprise deployments have been integrated with customer-managed identity providers, including organisations running an internal IdP alongside Entra.onedroid.ai/llms-full.txt (sources/llms-full.txt L4736-4738 (PAGE /security; under "Standards, not a bespoke scheme."))
    onedroid.ai/security.md (sources/security.md L127-129 (under "Standards, not a bespoke scheme."))
    Customer-IdP integrations are stated without a named deployment or artefact; the open repo has no match.
    C-231The signed-out application page was reworked to state its purpose up front and to render fully without JavaScript as part of that review (the application itself at synapse.onedroid.ai needs JavaScript).onedroid.ai/llms-full.txt (sources/llms-full.txt L6292-6295 (PAGE /changelog; under "Google OAuth application verification completed"))
    onedroid.ai/changelog.md (sources/changelog.md L183-186 (under "Google OAuth application verification completed"))
    The rework is stated in the changelog only; a GET of synapse.onedroid.ai returns a JavaScript-only shell, so no page rendering without JavaScript was found to back it.
    C-234Three comparisons · a written prediction before every run · no names [...] 17 October 2026 · 109 hidden checks · 3 runs per setup [...] 16 October 2026 · 13 hidden checks · 3 runs per setup [...] 7 October 2026 · 7 checks · two certified runs the same day [...] 5 and 6 October 2026 · 26 checks · three runs per systemonedroid.ai/llms-full.txt (sources/llms-full.txt L91-3171 (PAGE /; under "Three ways to test code you did not write"))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L3135-3171 (PAGE /case-studies; under "Three ways to test code you did not write"))
    The card dates and counts match the study pages, and the FINOS seal tx confirms its 7 October date, but the run records behind the counts are not public, so the list as a whole rests on the publisher's own pages.
    C-239Continuously. Once it passes, it goes on a schedule, every 15 minutes at the shortest, and one page answers “Is it OK?”.onedroid.ai/llms-full.txt (sources/llms-full.txt L112 (PAGE /; under "Same modernization, three setups: what the tester checks decides whether it helps"))Scheduling lives in the closed control plane (the skill only names author_set_schedule); the 15-minute floor and the 'Is it OK?' page were not found on a public surface.
    C-245The scenarios become regression guards; the next feature gets a fresh withheld set.onedroid.ai/llms-full.txt (sources/llms-full.txt L196-197 (PAGE /; under "The verdict, with evidence"))A process statement about reusing scenarios and withholding a fresh set; nothing public shows it applied.
    C-247OneDroid Engram is versioned knowledge and memory in a Postgres you control with your own database, or one OneDroid hosts — hybrid vector, keyword and graph search, git-style history, diff and restore.onedroid.ai/llms-full.txt (sources/llms-full.txt L232-233 (PAGE /; under "It brings its own run layer."))The Engram docs restate hybrid retrieval and bring-your-own Postgres; graph search and history, diff and restore were not shown by any public artefact.
    C-249The scenarios are written once per feature, before the build, and run as often as you need.onedroid.ai/llms-full.txt (sources/llms-full.txt L2783 (PAGE /argus; under "Not a test suite to maintain"))
    onedroid.ai/argus.md (sources/argus.md L47 (under "Not a test suite to maintain"))
    docs.onedroid.ai/argus (sources/argus-docs.html L75-76 (under "Is it end-to-end testing?"))
    A process statement about when scenarios are written; nothing public shows it applied.
    C-251Coding agents now build a feature for a few dollarsonedroid.ai/llms-full.txt (sources/llms-full.txt L2799-2800 (PAGE /argus; under "Testing was the part teams could skip. Not any more."))
    onedroid.ai/argus.md (sources/argus.md L63-64 (under "Testing was the part teams could skip. Not any more."))
    docs.onedroid.ai/argus (sources/argus-docs.html L88 (under "Why it exists"))
    The figure rests on the publisher's own case study ($1.01 to $2.78 a run) without public raw cost records.
    C-252No test proves code free of bugs. Different testers catch different things, and a model can give a different answer on each run.onedroid.ai/llms-full.txt (sources/llms-full.txt L2808-2809 (PAGE /argus; under "Testing was the part teams could skip. Not any more."))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L5276-5277 (PAGE /qa-teams; under "Not proof of no bugs"))
    onedroid.ai/argus.md (sources/argus.md L72-73 (under "Testing was the part teams could skip. Not any more."))
    A general disclaimer about testing that no artefact can back; it is restated on /argus and /qa-teams.
    C-259On 5 and 6 October it ran on two releases of an open source ticketing system, on one laptop. It found six differences between the releases.onedroid.ai/llms-full.txt (sources/llms-full.txt L2908 (PAGE /argus; under "One sealed set, two systems."))
    onedroid.ai/argus.md (sources/argus.md L172 (under "One sealed set, two systems."))
    The dates and the six differences rest on the case-study page; no public run record, hash or chain reference for the OTRS runs was found.
    C-265A keyword search of the change log did not find three of them.onedroid.ai/llms-full.txt (sources/llms-full.txt L3197 (PAGE /case-study-otrs-znuny; under "Six differences between two releases of a ticketing system"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L20 (under "Six differences between two releases of a ticketing system"))
    The keyword-search finding is the publisher's own account with no published search record.
    C-266The most important one was a fault in the OLD release: the new one fixes it.onedroid.ai/llms-full.txt (sources/llms-full.txt L3198 (PAGE /case-study-otrs-znuny; under "Six differences between two releases of a ticketing system"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L21 (under "Six differences between two releases of a ticketing system"))
    The emoji fault in OTRS 6.0.30 rests on the case-study table (U+FFFD back on OTRS, unchanged on Znuny) without a public run record.
    C-267An engineer on our team and his coding agent took two releases of an open source ticketing system and ran one sealed set of checks against both.onedroid.ai/llms-full.txt (sources/llms-full.txt L3208 (PAGE /case-study-otrs-znuny; under "In short"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L31 (under "In short"))
    Who ran the OTRS/Znuny study and that one sealed set ran against both is stated only; no set hash, anchor or run record for these sets is published.
    C-268Both from the official release archives with pinned checksums, on the same base imageonedroid.ai/llms-full.txt (sources/llms-full.txt L3236 (PAGE /case-study-otrs-znuny; under "The setup"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L59 (under "The setup"))
    The build method is stated only; no Dockerfile, compose file or list of pinned archive checksums is published.
    C-269One Windows laptop, Docker Compose, two OneDroid Argus executors side by side, no clusteronedroid.ai/llms-full.txt (sources/llms-full.txt L3242 (PAGE /case-study-otrs-znuny; under "The setup"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L65 (under "The setup"))
    The host setup is stated only; the runner README says a Docker Compose path exists but its onboarding kit is not shipped, and no compose file or executor record for this study is published.
    C-270Set 1, 5 October, 16 checks. [...] Set 2, 6 October, 26 checks.onedroid.ai/llms-full.txt (sources/llms-full.txt L3252-3254 (PAGE /case-study-otrs-znuny; under "What was checked"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L75-77 (under "What was checked"))
    Set sizes and dates appear in the text and the result table of the same page, but no sealed-set hash or ledger record for either set is published.
    C-271Checks | 16 | 26 | [...] Runs per system | 3 | 3 | Cells identical | 16 | 23 | Cells that differ | 4 | 9 | Cells that read as noise | 1 | 0 | Not measured | 0 | 0 | Could not run | 0 | 0onedroid.ai/llms-full.txt (sources/llms-full.txt L3264-3278 (PAGE /case-study-otrs-znuny; under "The result"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L87-101 (under "The result"))
    The cell states (identical, differs, noise, not_measured, could_not_run) exist in the compare code, but the counts are the tool's reported numbers with no comparison output, hash or anchor published.
    C-272Before the comparison ran, each set was sealed and its hash was recorded on a ledger, so a set could not be changed after anyone had seen a result.onedroid.ai/llms-full.txt (sources/llms-full.txt L3258 (PAGE /case-study-otrs-znuny; under "What was checked"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L81 (under "What was checked"))
    Sealing and anchoring exist as a product mechanism, but this study quotes no transaction, block or set hash for its two sets.
    C-273One thing the seal does not show: the checks were rehearsed on both systems before sealing.onedroid.ai/llms-full.txt (sources/llms-full.txt L3258 (PAGE /case-study-otrs-znuny; under "What was checked"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L81 (under "What was checked"))
    A disclosure about the authors' own process (checks rehearsed before sealing); nothing beyond the statement backs it.
    C-275The old URL path `/otrs/index.pl` | Answers | Answers 404 | Fixed check, 0 of 3 runs on Znunyonedroid.ai/llms-full.txt (sources/llms-full.txt L3288 (PAGE /case-study-otrs-znuny; under "The six differences"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L111 (under "The six differences"))
    The /otrs/index.pl fixed-check result (0 of 3 on Znuny) is stated with no run output or revealed check; repo-grep found nothing.
    C-276An emoji in an article | Comes back as U+FFFD, the replacement character | Comes back unchanged | Fixed check: 0 of 3 on OTRS, 3 of 3 on Znunyonedroid.ai/llms-full.txt (sources/llms-full.txt L3290 (PAGE /case-study-otrs-znuny; under "The six differences"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L113 (under "The six differences"))
    The emoji U+FFFD per-run results are stated with no recorded outputs or revealed check published.
    C-277`CommunicationChannel` on every article | Not there | There | Measured, on three different reads: plain, with an attachment, with a customer sessiononedroid.ai/llms-full.txt (sources/llms-full.txt L3292 (PAGE /case-study-otrs-znuny; under "The six differences"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L115 (under "The six differences"))
    The CommunicationChannel field difference is stated with no recorded outputs published; repo-grep found nothing.
    C-278The `DynamicField` list on a ticket | The list, with empty values | No list | Measured. The cause was not foundonedroid.ai/llms-full.txt (sources/llms-full.txt L3294 (PAGE /case-study-otrs-znuny; under "The six differences"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L117 (under "The six differences"))
    The DynamicField list difference is stated with no recorded outputs published.
    C-279Ticket history after create, close and reopen | 11 entries | 13 entries | Measured. The extra entries are `SetPendingTime`, written at a state changeonedroid.ai/llms-full.txt (sources/llms-full.txt L3296 (PAGE /case-study-otrs-znuny; under "The six differences"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L119 (under "The six differences"))
    The 11 versus 13 history entries and the SetPendingTime cause are stated with no recorded output published.
    C-280Keys in the session data | 1 key only here | 7 keys only here | Measured with the values masked. 26 keys are commononedroid.ai/llms-full.txt (sources/llms-full.txt L3298 (PAGE /case-study-otrs-znuny; under "The six differences"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L121 (under "The six differences"))
    The 1/7/26 session-key counts are stated on the page twice but no masked output is published.
    C-281Six error and refusal paths answer the same on both.onedroid.ai/llms-full.txt (sources/llms-full.txt L3308 (PAGE /case-study-otrs-znuny; under "What was the same"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L131 (under "What was the same"))
    Stated only; no per-check run record for the six error and refusal paths is published.
    C-282Customer permissions: a customer is refused a ticket that is not theirs, and can raise the priority of their own.onedroid.ai/llms-full.txt (sources/llms-full.txt L3310 (PAGE /case-study-otrs-znuny; under "What was the same"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L133 (under "What was the same"))
    Stated only; no per-check run record for the customer permission checks is published.
    C-283Pending reminder, close, and a queue move with owner and lock. [...] Ticket search and the entry pages.onedroid.ai/llms-full.txt (sources/llms-full.txt L3312-3314 (PAGE /case-study-otrs-znuny; under "What was the same"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L135-137 (under "What was the same"))
    Stated only; no per-check run record for the workflow, search and entry-page checks is published.
    C-284In the first set the session-data check compared raw values. The reference gave three different outputs in three runs, so OneDroid Argus marked the cell as noise and did not call it a difference.onedroid.ai/llms-full.txt (sources/llms-full.txt L3322 (PAGE /case-study-otrs-znuny; under "What the noise taught us"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L145 (under "What the noise taught us"))
    A `noise` cell state exists in the compare code, but this run's three reference outputs and the noise verdict for the cell are not published.
    C-285Compare is in early use and ran here against a development control planeonedroid.ai/llms-full.txt (sources/llms-full.txt L3346 (PAGE /case-study-otrs-znuny; under "What this test did not cover"))
    onedroid.ai/case-study-otrs-znuny.md (sources/case-study-otrs-znuny.md L169 (under "What this test did not cover"))
    A disclosure about the environment (early use, development control plane), stated twice on the page; nothing beyond the statement backs it.
    C-286The demo's own manifests, unchanged, on a shared Azure Kubernetes cluster in Poland Central.onedroid.ai/llms-full.txt (sources/llms-full.txt L3432 (PAGE /case-study-finos-calm; under "The setup"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L59 (under "The setup"))
    The cluster and the unchanged manifests are stated only; the manifests are in a private bundle (sha256:37e5df71…) shown on request, not published.
    C-290An earlier certified run of the same seven checks the same day gave the same verdict.onedroid.ai/llms-full.txt (sources/llms-full.txt L3480 (PAGE /case-study-finos-calm; under "The result"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L107 (under "The result"))
    The earlier certified run is mentioned twice on the page, but no transaction, block or certificate for it is quoted.
    C-291We did not trace the cause in the demo's code, and we have not reported it to the demo's authors.onedroid.ai/llms-full.txt (sources/llms-full.txt L3484 (PAGE /case-study-finos-calm; under "The two checks that failed"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L111 (under "The two checks that failed"))
    A conduct disclosure (cause not traced, not reported to the authors); nothing beyond the statement backs it.
    C-295The seal proves priority: the checks existed, unchanged, before the result. It does not make their authors blind to the system. The same team wrote the checks and ran them.onedroid.ai/llms-full.txt (sources/llms-full.txt L3512 (PAGE /case-study-finos-calm; under "How anyone can check it later"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L139 (under "How anyone can check it later"))
    The limit of the seal is stated on the page and echoed in the runner README (an anchor proves existence no later than its block, not truth); the authorship statement has nothing beyond the claim.
    C-298Claude Code on Claude Opus 5.5, high effort.onedroid.ai/llms-full.txt (sources/llms-full.txt L3616 (PAGE /case-study-claude-code-alone-vs-tester; under "The setup"))
    onedroid.ai/case-study-claude-code-alone-vs-tester.md (sources/case-study-claude-code-alone-vs-tester.md L53 (under "The setup"))
    The model and effort setting are stated only; no transcript or run log is published.
    C-300The build checks ran together with 37 older checks of the same app, as regression checks.onedroid.ai/llms-full.txt (sources/llms-full.txt L3620 (PAGE /case-study-claude-code-alone-vs-tester; under "The setup"))
    onedroid.ai/case-study-claude-code-alone-vs-tester.md (sources/case-study-claude-code-alone-vs-tester.md L57 (under "The setup"))
    The 37 older regression checks are stated three times on the page, with no check list or run record published.
    C-30113 hidden checks, stored in OneDroid Argus only after the last run, so neither setup saw them, not even as names.onedroid.ai/llms-full.txt (sources/llms-full.txt L3624 (PAGE /case-study-claude-code-alone-vs-tester; under "The setup"))
    onedroid.ai/case-study-claude-code-alone-vs-tester.md (sources/case-study-claude-code-alone-vs-tester.md L61 (under "The setup"))
    When the hidden checks were stored is stated only; no store timestamp, seal or anchor for them is published.
    C-305On an easy ticket, an independent tester buys no extra quality and costs more in total.onedroid.ai/llms-full.txt (sources/llms-full.txt L3682 (PAGE /case-study-claude-code-alone-vs-tester; under "What we take from it"))
    onedroid.ai/case-study-claude-code-alone-vs-tester.md (sources/case-study-claude-code-alone-vs-tester.md L119 (under "What we take from it"))
    The conclusion follows from the page's own table (13/13 both, $2.78 against $1.97), but the numbers rest on a private bundle (sha256:759cf275…), not a published record.
    C-306The ticket was too easy to separate quality.onedroid.ai/llms.txt (sources/llms.txt L62 (under "Docs"))The same judgement is repeated on the case-study page; both setups' 13/13 result is reported with no judge outputs published.
    C-312All three B runs changed one numerical detail that the original handles approximately, which cost 22 to 25 of the 109 hidden checks.onedroid.ai/llms-full.txt (sources/llms-full.txt L3778 (PAGE /case-study-modernization-three-setups; under "What the numbers say"))
    onedroid.ai/case-study-modernization-three-setups.md (sources/case-study-modernization-three-setups.md L87 (under "What the numbers say"))
    The 22-25 hidden-check loss is consistent with B's mean of 81 against 104 for A, but no per-run judge output is published.
    C-314B′ cost 41% less than A, because a coding agent that only implements runs no tests and no subagents.onedroid.ai/llms-full.txt (sources/llms-full.txt L3782 (PAGE /case-study-modernization-three-setups; under "What the numbers say"))
    onedroid.ai/case-study-modernization-three-setups.md (sources/case-study-modernization-three-setups.md L91 (under "What the numbers say"))
    The 41% and the 0/0 test-run and subagent counts come from the page's own table, with no transcripts or cost records published.
    C-317The tester's checks came from earlier work on the same library and can be reused. Writing them is not in the cost per run.onedroid.ai/llms-full.txt (sources/llms-full.txt L3800 (PAGE /case-study-modernization-three-setups; under "What this experiment did not cover"))
    onedroid.ai/case-study-modernization-three-setups.md (sources/case-study-modernization-three-setups.md L109 (under "What this experiment did not cover"))
    A disclosure about how cost was counted; nothing beyond the statement backs it.
    C-334OneDroid Argus installs Synapse. You can also run it on its own.onedroid.ai/llms-full.txt (sources/llms-full.txt L3919 (PAGE /synapse; under "OneDroid Synapse"))The open-source runner repo has no Synapse install path (git grep `synapse` finds nothing); nothing beyond the statement backs it.
    C-335Free on the hosted shape, whatever the size of your team. No credit card.onedroid.ai/llms-full.txt (sources/llms-full.txt L3928 (PAGE /synapse; under "OneDroid Synapse"))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L4372 (PAGE /what-is-a-governed-mcp-gateway; under "Try it against your own setup."))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L4815 (PAGE /engram; under "Portable, persistent context"))
    The free hosted tier is stated on several pages; no published price list entry or sign-up flow without a card is recorded.
    C-337Admins decide which tools each hub, role, and agent can reach — as curated packs, not one-off allowlists. Users connect their own credentials once; agents never handle raw tokens.onedroid.ai/llms-full.txt (sources/llms-full.txt L3984-3986 (PAGE /synapse; under "Govern"))Packs and credential custody are described on the page and in screenshots only; the gateway's code or policy schema is not public.
    C-338Add or swap upstreams without touching a single agent config.onedroid.ai/llms-full.txt (sources/llms-full.txt L4002-4003 (PAGE /synapse; under "One URL for every agent"))Stated only; no public artefact shows upstream changes leaving agent configs untouched.
    C-339Separate workspaces for teams and projects, each with its own tool surface, members, and defaults.onedroid.ai/llms-full.txt (sources/llms-full.txt L4022-4023 (PAGE /synapse; under "Hubs and roles"))Per-hub URLs (/hub/<slug>/mcp) suggest separate hubs, but their per-hub members and defaults are described only.
    C-340With your own database, your Engram content lives in a Postgres you own; on the hosted gateway OneDroid still holds encrypted credentials and sees tool-call metadata; self-hosted, OneDroid sees nothing at runtime.onedroid.ai/llms-full.txt (sources/llms-full.txt L4051 (PAGE /synapse; under "Your data lives in your database."))The data-visibility split per topology is repeated on /security, with no public code, architecture or audit backing it.
    C-341Your database, your jurisdiction, your retention policy. Governance data never has to leave your infrastructure.onedroid.ai/llms-full.txt (sources/llms-full.txt L4055-4056 (PAGE /synapse; under "Data governance"))Stated only; no self-host package or data-flow artefact is published.
    C-342Self-host OneDroid Synapse and credentials and the event log stay inside your own boundary as well.onedroid.ai/llms-full.txt (sources/llms-full.txt L4180 (PAGE /synapse; under "Where does my data live with OneDroid?"))Repeated on /security; no self-host distribution or deployment artefact is published.
    C-343The animation further up this page is illustrative and labelled as such. These are screenshots of OneDroid Synapse itself.onedroid.ai/llms-full.txt (sources/llms-full.txt L4144-4145 (PAGE /synapse; under "Not a mockup."))The animation is labelled 'illustrative' on the page, but whether the screenshots are of the live product cannot be checked from them.
    C-344Works with any LLM, any MCP client, any infrastructure.onedroid.ai/llms-full.txt (sources/llms-full.txt L4132 (PAGE /synapse; under "Portability"))Universal compatibility is stated; the live MCP endpoint backs the MCP-client part only, and nothing backs 'any LLM' or 'any infrastructure'.
    C-345State, credentials, audit records, and configuration live where you deploy them: inside your boundary when self-hosted.onedroid.ai/llms-full.txt (sources/llms-full.txt L4126-4127 (PAGE /synapse; under "Ownership"))Repeated on /security; no self-host artefact shows where state lives.
    C-346Gateway: one extra hop. This one genuinely favours config files.onedroid.ai/llms-full.txt (sources/llms-full.txt L4294 (PAGE /what-is-a-governed-mcp-gateway; under "Latency"))An analytical statement about gateways in general; no measurement of the added hop is published.
    C-347Gateway: it is shared infrastructure — if it is down, everything behind it is unreachable.onedroid.ai/llms-full.txt (sources/llms-full.txt L4300-4301 (PAGE /what-is-a-governed-mcp-gateway; under "Failure blast radius"))An analytical statement about shared infrastructure; nothing beyond the statement backs it.
    C-348For the record, OneDroid Synapse answers these as: per tool; self-hostable so we can see nothing; the event log is in the free tier and will not be gated; your data can live in your own Postgres; any MCP client and any provider; and if you leave, the data was already in your database.onedroid.ai/llms-full.txt (sources/llms-full.txt L4360-4363 (PAGE /what-is-a-governed-mcp-gateway; under "Exit cost"))The six answers repeat claims made elsewhere on /synapse; no self-host package, pricing record or BYO-Postgres artefact is published, and repo-grep found nothing.
    C-351Engram is versioned, permissioned context in a Postgres you control with your own database, or one OneDroid hosts, searched by vector, keyword and graph at once, and reachable by any MCP client behind any model.onedroid.ai/llms-full.txt (sources/llms-full.txt L4472-4473 (PAGE /quickstart; under "Give your agents memory that outlives the session."))Engram's storage and hybrid search are described only; the Engram server code is not public and no endpoint artefact backs the capability list.
    C-362The control plane signs and anchors what the runner reports; it does not re-run the checks.onedroid.ai/llms-full.txt (sources/llms-full.txt L4643-4644 (PAGE /security; under "Where the verifier runs, and where it never does."))
    raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L26-27 (under "What a OneDroid Argus certificate proves today, and what it does not"))
    onedroid.ai/security.md (sources/security.md L34-35 (under "Where the verifier runs, and where it never does."))
    The runner README and /security both say this, but the control plane's code is not public, so nothing beyond the statement shows it does not re-run checks.
    C-363OneDroid Synapse runs on OneDroid infrastructure. Fastest to start. We operate the gateway, hold encrypted credentials, and can see tool-call metadata.onedroid.ai/llms-full.txt (sources/llms-full.txt L4654-4655 (PAGE /security; under "Hosted"))
    onedroid.ai/security.md (sources/security.md L45-46 (under "Hosted"))
    The hosted topology's custody and visibility are stated on /security and the Synapse FAQ, with no public artefact.
    C-364The gateway is ours; OneDroid Engram's data lives in a Postgres you own and control. Documents, embeddings, history and permissions live in your database; the library registry (which libraries exist and who belongs to them) is read from OneDroid's platform database, and embeddings are computed by OpenAI before they are stored.onedroid.ai/llms-full.txt (sources/llms-full.txt L4659-4660 (PAGE /security; under "Hosted + your database"))
    onedroid.ai/security.md (sources/security.md L50-51 (under "Hosted + your database"))
    The split between the customer's Postgres, the platform registry and OpenAI embeddings is stated only; the Engram code and data flow are not public.
    C-365OneDroid Synapse runs on your infrastructure, cloud or on-premises. Credentials and the event log stay inside your boundary. We can see nothing at runtime.onedroid.ai/llms-full.txt (sources/llms-full.txt L4664-4665 (PAGE /security; under "Self-hosted (enterprise)"))
    onedroid.ai/security.md (sources/security.md L55-56 (under "Self-hosted (enterprise)"))
    The self-hosted topology is stated only; no self-host distribution is published, and repo-grep for on-premises found nothing.
    C-367Everything above, minus your OneDroid Engram content: documents, chunks, embeddings, revision history and permissions are rows in your Postgres (the library registry, which libraries exist and who belongs to them, stays in OneDroid's platform database). You can read them with SQL and back them up with your own tooling.onedroid.ai/llms-full.txt (sources/llms-full.txt L4683-4685 (PAGE /security; under "Bring-your-own database"))
    onedroid.ai/security.md (sources/security.md L74-76 (under "Bring-your-own database"))
    A bring-your-own-database option exists in the hosted app, but which tables hold documents, chunks, embeddings, revisions and permissions, and that the registry stays on the platform, is stated only; a schema or migration would be needed.
    C-369Enterprise adds audit evidence packs for compliance reviews.onedroid.ai/llms-full.txt (sources/llms-full.txt L4713 (PAGE /security; under "Append-only audit log"))
    onedroid.ai/security.md (sources/security.md L104 (under "Append-only audit log"))
    Audit evidence packs are an Enterprise offering stated on /security and /pricing; no sample pack or pack format is published.
    C-372With your own database, every document, embedding and revision is a row in a Postgres you control.onedroid.ai/llms-full.txt (sources/llms-full.txt L4810 (PAGE /engram; under "Portable, persistent context"))Where each document, embedding and revision is stored under BYO is stated only; a published schema or migration set would be needed.
    C-373Hybrid retrieval — semantic and keyword together — over a typed object graph, so relationships like "supersedes" or "references" are explicit and traversable rather than inferred from similarity.onedroid.ai/llms-full.txt (sources/llms-full.txt L4850-4852 (PAGE /engram; under "Rich, not just a vector blob"))The Engram client has a library edges route, but hybrid semantic-plus-keyword retrieval and typed 'supersedes'/'references' links are not shown by any public artefact.
    C-376Move from one client or provider to another, run two side by side, or test an open-source model — the tools-and-context layer doesn't move.onedroid.ai/llms-full.txt (sources/llms-full.txt L4968-4970 (PAGE /financial-services; under "Change the model without a migration"))
    onedroid.ai/financial-services.md (sources/financial-services.md L54-56 (under "Change the model without a migration"))
    Model portability is stated only; a demonstrated switch of client or model against the same hub would be needed.
    C-377You don't assemble the compliance story afterwards; running the platform produces it.onedroid.ai/llms-full.txt (sources/llms-full.txt L4975-4976 (PAGE /financial-services; under "Evidence as a by-product"))
    onedroid.ai/financial-services.md (sources/financial-services.md L61-62 (under "Evidence as a by-product"))
    That operating the platform produces the compliance record is stated only; activity logging is a fail-open hub switch, and no sample compliance output is published.
    C-378Because OneDroid Synapse sits in the call path from day one, the answers exist before anyone asks.onedroid.ai/llms-full.txt (sources/llms-full.txt L4985-4986 (PAGE /financial-services; under "Most AI programmes don't fail on capability."))
    onedroid.ai/financial-services.md (sources/financial-services.md L71-72 (under "Most AI programmes don't fail on capability."))
    A positioning statement about Synapse in the call path; no published record shows such answers being produced.
    C-379This produces a tamper-evident one a third party can check.onedroid.ai/llms-full.txt (sources/llms-full.txt L4996-4997 (PAGE /financial-services; under "Supervision and records"))
    onedroid.ai/financial-services.md (sources/financial-services.md L82-83 (under "Supervision and records"))
    A hash-chain audit switch exists in the hosted app, but no tool or procedure lets a third party check a Synapse record; one would be needed.
    C-380Deployed in your own cloud, on your own Postgres.onedroid.ai/llms-full.txt (sources/llms-full.txt L5001 (PAGE /financial-services; under "Operations"))
    onedroid.ai/financial-services.md (sources/financial-services.md L87 (under "Operations"))
    Customer-cloud deployment on customer Postgres is an Enterprise statement; no deployment manifest or reference install is published.
    C-381Model inference is an outbound call — for us and for everyone using commercial AI. Your data comes to rest inside your boundary; the inference call that transits is gated, scanned and logged.onedroid.ai/llms-full.txt (sources/llms-full.txt L5114-5116 (PAGE /financial-services; under "Not "nothing leaves""))
    onedroid.ai/financial-services.md (sources/financial-services.md L200-202 (under "Not "nothing leaves""))
    That the inference call is gated, scanned and logged is stated only; no gateway code or log sample is public.
    C-382OneDroid Synapse transports, authorises and logs tool calls. It does not vouch for an agent's judgement, and no product does.onedroid.ai/llms-full.txt (sources/llms-full.txt L5121-5122 (PAGE /financial-services; under "Not a safety guarantee"))
    onedroid.ai/financial-services.md (sources/financial-services.md L207-208 (under "Not a safety guarantee"))
    A scope disclaimer; the grant and activity-log routes show authorisation and logging surfaces, but the disclaimer itself has no artefact.
    C-383Any one of those is a sprint. All of them, tested, and defensible in front of an examiner, is two quarters of your platform team not building product.onedroid.ai/llms-full.txt (sources/llms-full.txt L5055-5056 (PAGE /financial-services; under ""What happens when we leave?""))
    onedroid.ai/financial-services.md (sources/financial-services.md L141-142 (under ""What happens when we leave?""))
    A 'two quarters' effort estimate with no published basis (no task breakdown or measured effort).
    C-384A set goes on a schedule once a hand-run is green, or every red in it is one you have read and decided to keep.onedroid.ai/llms-full.txt (sources/llms-full.txt L5250-5251 (PAGE /qa-teams; under "Which reds you keep"))The skill names a schedule tool for sealed sets, but the gate that a set must be green or have only kept reds before scheduling sits in the hosted control plane and is not shown.
    C-385What we sell alongside the kits is help making them work inside your organisation; the products themselves are on /pricing.onedroid.ai/llms-full.txt (sources/llms-full.txt L5325 (PAGE /dark-factory; under "Autonomous delivery"))/pricing exists and lists the products, but what is sold alongside the kits (implementation help) is a business statement with no artefact.
    C-386Works with any agent runtime that can follow instructions and run commands — no proprietary orchestrator required.onedroid.ai/llms-full.txt (sources/llms-full.txt L5373-5374 (PAGE /dark-factory; under "Harness-agnostic"))The kits carry a Codex hook contract besides Claude Code hooks, but 'any agent runtime' is a universal claim no artefact can back.
    C-388Everything we sell comes down to one thing: control over what your agents can reach, proof of what they did, and a verdict on what they built.onedroid.ai/llms-full.txt (sources/llms-full.txt L5762-5763 (PAGE /enterprise; under "Fully govern your AI agent"))A positioning summary of the offer; nothing beyond the statement backs it.
    C-389Governance data never leaves your perimeter.onedroid.ai/llms-full.txt (sources/llms-full.txt L5791 (PAGE /enterprise; under "Your infrastructure"))An Enterprise data-residency statement; a deployment design or network policy showing no egress of governance data would be needed.
    C-391Enterprise is the whole stack inside your perimeteronedroid.ai/llms-full.txt (sources/llms-full.txt L5847 (PAGE /enterprise; under "OneDroid Argus, Synapse and Engram, on your infrastructure."))An Enterprise packaging statement; no on-premises install artefact for the whole stack is published.
    C-399Publishes the reveal: trigger, expectation, evidence hash, verdictonedroid.ai/llms-full.txt (sources/llms-full.txt L5936 (PAGE /pricing; under "The OneDroid Argus control plane"))
    onedroid.ai/pricing.md (sources/pricing.md L51 (under "The OneDroid Argus control plane"))
    The open CLI reads and replays a reveal document, but no published reveal (trigger, expectation, evidence hash, verdict) was found on the saved surfaces; one would be needed.
    C-403Deployed on your infrastructure — cloud or on-premisesonedroid.ai/llms-full.txt (sources/llms-full.txt L5980 (PAGE /pricing; under "Enterprise"))
    onedroid.ai/pricing.md (sources/pricing.md L95 (under "Enterprise"))
    An Enterprise deployment statement; no install or reference deployment for customer infrastructure is published.
    C-404Org-wide policy enforcement: deny rules, grants, data filtersonedroid.ai/llms-full.txt (sources/llms-full.txt L5982 (PAGE /pricing; under "Enterprise"))
    onedroid.ai/pricing.md (sources/pricing.md L97 (under "Enterprise"))
    Per-hub grants and toggles exist on the hosted shape, but org-wide enforcement, deny rules and data filters for Enterprise are not shown.
    C-405PII / PHI screening set up for your regulated workloads (the switch itself is on every hub)onedroid.ai/llms-full.txt (sources/llms-full.txt L5984 (PAGE /pricing; under "Enterprise"))
    onedroid.ai/pricing.md (sources/pricing.md L99 (under "Enterprise"))
    The per-hub PHI Detection switch exists in the hosted app, which backs the parenthesis; the Enterprise set-up for regulated workloads is a service statement.
    C-406Audit evidence packs for compliance reviewsonedroid.ai/llms-full.txt (sources/llms-full.txt L5986 (PAGE /pricing; under "Enterprise"))
    onedroid.ai/pricing.md (sources/pricing.md L101 (under "Enterprise"))
    Audit evidence packs are listed only; no sample or format is published.
    C-407Implementation support from the team that builds itonedroid.ai/llms-full.txt (sources/llms-full.txt L5988 (PAGE /pricing; under "Enterprise"))
    onedroid.ai/pricing.md (sources/pricing.md L103 (under "Enterprise"))
    A service commitment with no artefact.
    C-408Free covers any team, whatever its size, using OneDroid's hosted OneDroid Synapse and OneDroid Engram, including the event log.onedroid.ai/llms-full.txt (sources/llms-full.txt L5995-5996 (PAGE /pricing; under "Enterprise"))
    onedroid.ai/pricing.md (sources/pricing.md L110-111 (under "Enterprise"))
    A pricing statement; no plan list or billing artefact shows the free scope.
    C-409Enterprise begins when you need it on your own infrastructure, org-wide policy enforcement, SSO, or audit evidence packs for a compliance review.onedroid.ai/llms-full.txt (sources/llms-full.txt L5996-5998 (PAGE /pricing; under "Enterprise"))
    onedroid.ai/pricing.md (sources/pricing.md L111-113 (under "Enterprise"))
    A pricing boundary statement with no artefact beyond the page.
    C-410The free tier is genuinely free while we build in the open.onedroid.ai/llms-full.txt (sources/llms-full.txt L6002 (PAGE /pricing; under "Enterprise"))
    onedroid.ai/pricing.md (sources/pricing.md L117 (under "Enterprise"))
    A conduct statement about pricing; nothing beyond the statement backs it.
    C-412OneDroid is the product line that came out of that work.onedroid.ai/llms-full.txt (sources/llms-full.txt L6056 (PAGE /about; under "OneDroid is built by Providentia Worldwide."))An origin statement; no record of the prior Providentia work is published.
    C-413Every component here exists because a real deployment needed it: the gateway because credentials were being pasted into config files, the memory because knowledge kept dying with the context window, the audit log because someone had to answer a compliance question about what an agent had touched.onedroid.ai/llms-full.txt (sources/llms-full.txt L6056-6060 (PAGE /about; under "OneDroid is built by Providentia Worldwide."))An origin story for each component; no deployment record or incident is published.
    C-416A run whose executor was restarted mid-run is now reported as failed, with the reason, instead of staying open.onedroid.ai/llms-full.txt (sources/llms-full.txt L6140-6141 (PAGE /changelog; under "OneDroid Argus 0.3.74: a check can listen on a message bus, and read the logs an environment already keeps"))
    onedroid.ai/changelog.md (sources/changelog.md L31-32 (under "OneDroid Argus 0.3.74: a check can listen on a message bus, and read the logs an environment already keeps"))
    The executor code says the control plane is the authority on a run in flight after a restart, but marking such a run failed with a reason happens in the hosted control plane and is not shown.
    C-4262026-10-20 ### The site now leads with what testing costsonedroid.ai/llms-full.txt (sources/llms-full.txt L6149-6151 (PAGE /changelog; under "OneDroid Argus 0.3.74: a check can listen on a message bus, and read the logs an environment already keeps"))
    onedroid.ai/changelog.md (sources/changelog.md L40-42 (under "OneDroid Argus 0.3.74: a check can listen on a message bus, and read the logs an environment already keeps"))
    A dated changelog entry with no artefact of the 20 October page; the current home page leads with in-situ testing and carries the cost section further down.
    C-431There is no sales queue and no contact-form robot.onedroid.ai/llms-full.txt (sources/llms-full.txt L6350 (PAGE /contact; under "Talk to the people"))The contact page gives a direct email, but the absence of a sales queue or form robot is stated only.
    C-432We collect the minimum needed to run a governed gateway: your account details, the credentials you choose to connect, a metadata log of tool calls and, if you use them, your Engram content and the server logs described below.onedroid.ai/llms-full.txt (sources/llms-full.txt L6418-6420 (PAGE /privacy; under "The short version"))
    onedroid.ai/privacy.md (sources/privacy.md L27-29 (under "The short version"))
    A data-minimisation statement; a data inventory or schema would be needed.
    C-433OneDroid Synapse is designed so that agents receive tool results, not the underlying tokens — raw credentials are held by the gateway and are not exposed to connected agents.onedroid.ai/llms-full.txt (sources/llms-full.txt L6440-6442 (PAGE /privacy; under "What we collect"))
    onedroid.ai/privacy.md (sources/privacy.md L49-51 (under "What we collect"))
    That agents receive results, not tokens, is a gateway property whose code is not public; a public implementation or test would be needed.
    C-434If you save Google-sourced content into OneDroid Engram, search embeddings for it are computed by our embedding provider (OpenAI) under terms that prohibit training on your data.onedroid.ai/llms-full.txt (sources/llms-full.txt L6497-6499 (PAGE /privacy; under "Google user data"))
    onedroid.ai/privacy.md (sources/privacy.md L106-108 (under "Google user data"))
    The embedding provider and its no-training terms are stated only; no subprocessor list or DPA reference is published.
    C-435On disconnection or account deletion, stored Google tokens are deleted.onedroid.ai/llms-full.txt (sources/llms-full.txt L6502-6503 (PAGE /privacy; under "Google user data"))
    onedroid.ai/privacy.md (sources/privacy.md L111-112 (under "Google user data"))
    A token-deletion commitment stated in the policy only.
    C-436OneDroid's use and transfer of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements.onedroid.ai/llms-full.txt (sources/llms-full.txt L6505-6507 (PAGE /privacy; under "Google user data"))
    onedroid.ai/privacy.md (sources/privacy.md L114-116 (under "Google user data"))
    Limited Use adherence is stated, and the changelog's OAuth verification entry is the publisher's own statement, not a Google artefact.
    C-437We do not sell personal data, and we do not share it with third parties for their marketing.onedroid.ai/llms-full.txt (sources/llms-full.txt L6522-6523 (PAGE /privacy; under "How we share data"))
    onedroid.ai/privacy.md (sources/privacy.md L131-132 (under "How we share data"))
    A no-sale commitment stated in the policy only.
    C-438We do not sell or share personal data as those terms are defined in the CCPA.onedroid.ai/llms-full.txt (sources/llms-full.txt L6557-6558 (PAGE /privacy; under "Where GDPR or CCPA applies"))
    onedroid.ai/privacy.md (sources/privacy.md L166-167 (under "Where GDPR or CCPA applies"))
    A CCPA no-sale/no-share statement in the policy only.
    C-439Effective date: 19 July 2026 · Last updated: 7 October 2026onedroid.ai/llms-full.txt (sources/llms-full.txt L6599 (PAGE /terms; under "Terms of Service"))The live terms page carries these dates and the changelog dates publication to 19 July, but both are the publisher's own statements; a version history of the terms would be needed for 'last updated'.
    C-440The Service is free on the hosted shape, whatever the size of your team; current usage limits, if any, are published on the pricing page.onedroid.ai/llms-full.txt (sources/llms-full.txt L6618-6620 (PAGE /terms; under "2. The Service"))A pricing term repeated on /terms and /pricing; nothing beyond the statements backs it.
    C-441You must be at least 16 years old to use the Service.onedroid.ai/llms-full.txt (sources/llms-full.txt L6635 (PAGE /terms; under "3. Your account"))An age requirement in the terms; no age gate in the sign-up flow is observable.
    C-442After termination we will make your platform-hosted data available for export for 30 days where practicable, then delete it.onedroid.ai/llms-full.txt (sources/llms-full.txt L6763-6764 (PAGE /terms; under "12. Suspension and termination"))A post-termination export commitment stated in the terms only.
    C-451Each tick queues a `build` run of your reviewed set, marked with the intent `monitor`.docs.onedroid.ai/argus.md (sources/argus-docs.md L100 (under "The two ways to start a run"))Scheduling and the run intent live in the hosted control plane; the public runner has no `monitor` run intent, so only the docs page states it; a control-plane run record showing intent monitor would be needed.
    C-452These pages describe Argus v0.3.72.docs.onedroid.ai/argus (sources/argus-docs.html L163 (under "What changed recently"))
    docs.onedroid.ai/argus.md (sources/argus-docs.md L114 (under "What changed recently"))
    No public release record ties the pages to v0.3.72: the open repo has no releases or tags and the release repo is private (quickstart L27); repo docs mention 0.3.72 only as the CLI that reads v5 certificates.
    C-467**Proof page** (v0.3.53): certified releases first, filters and search, and an **Anchored** filter on Runs.docs.onedroid.ai/argus.md (sources/argus-docs.md L237-238 (under "What changed recently"))The Proof page is part of the hosted web app, which is not in the public repository; only the changelog states it; a public view of the page would be needed.
    C-468**Where things moved in the web app** (v0.3.53): Proof was called Ledger.docs.onedroid.ai/argus.md (sources/argus-docs.md L257 (under "What changed recently"))The rename of Ledger to Proof concerns the hosted web app; nothing public beyond the changelog line shows it.
    C-479A builder never needs the cloud session, because a builder never calls a `cloud-*` command.docs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L142-143 (under "2. The two kinds of token"))The code defines a builder namespace as runner scope (types.go L79-89) but nothing shows that no builder path calls a cloud-* command; a command gate refusing cloud-* to a builder would be needed.
    C-481If that page approves the same code twice (a reload, a double click), it says it is approved (v0.3.52 or later)docs.onedroid.ai/argus-quickstart (sources/argus-quickstart.html L197-198 (under "3. Point the CLI at your control plane"))
    docs.onedroid.ai/argus-quickstart.md (sources/argus-quickstart.md L179-180 (under "3. Point the CLI at your control plane"))
    The approval page is part of the hosted control plane, not the public repo; only the quickstart states the idempotent approval.
    C-487So **a modified runner could report passes it did not earn**, and the certificate would still be anchored and would still verifyraw.githubusercontent.com/OneDro1d/argus-runner/main/README.md (sources/argus-runner-README.md L27-28 (under "What a OneDroid Argus certificate proves today, and what it does not"))verify.go's caveats say some fields are attested by the executor only, but no artefact shows a modified runner's unearned passes still verifying; that would need a demonstration run.
    C-516For years companies spent between 23% and 35% of their IT budget on quality assurance and testing (23% in 2013, 26% in 2014, 35% in 2015, 31% in 2016, 26% in 2017) [...] Testing took people and time, so it was the part that got cut.onedroid.ai/llms-full.txt (sources/llms-full.txt L2797-2799 (PAGE /argus; under "Testing was the part teams could skip. Not any more."))
    onedroid.ai/argus.md (sources/argus.md L61-63 (under "Testing was the part teams could skip. Not any more."))
    The yearly percentages are attributed to Capgemini's World Quality Report with no link, table or fetched copy of the reports; the cited reports would be needed.
    C-517Your agents connect to one URL instead of holding credentials for each serviceonedroid.ai/llms-full.txt (sources/llms-full.txt L227-228 (PAGE /; under "It brings its own run layer."))Synapse is hosted and not in a public repository; the one-URL gateway is only stated on the pages.
    C-518every tool call is authenticated, policy-checked, and logged while the hub's activity logging is on (fail-open: a logging failure never blocks a call), to an append-only audit log you ownonedroid.ai/llms-full.txt (sources/llms-full.txt L228-229 (PAGE /; under "It brings its own run layer."))
    onedroid.ai/llms.txt (sources/llms.txt L84 (under "Product"))
    Authentication, policy checks, fail-open logging and the append-only audit log are stated on several pages; no public Synapse code or sample log export backs them.
    C-519tool call is authenticated, policy-checked, and logged while the hub's activity logging is on (fail-open: a logging failure never blocks a call), to an audit log you ownonedroid.ai/llms-full.txt (sources/llms-full.txt L3909 (PAGE /synapse))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L3922-3923 (PAGE /synapse; under "OneDroid Synapse"))
    The same Synapse audit-log statement as C-518, restated with no public code, schema or sample export.
    C-520Portable, persistent, permissioned context for every AI application — versioned, searchable by any MCP client, stored in your own database or hosted by OneDroid.onedroid.ai/llms-full.txt (sources/llms-full.txt L4084-4085 (PAGE /synapse; under "OneDroid Engram"))Engram's versioned, searchable, self-hostable context is stated only; no public repo or public endpoint documentation backs it.
    C-521Yes, on the hosted shape, whatever the size of your team, including the event logonedroid.ai/llms-full.txt (sources/llms-full.txt L4172 (PAGE /synapse; under "Is OneDroid Synapse free?"))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L6012 (PAGE /pricing; under "Is OneDroid Synapse free?"))
    onedroid.ai/pricing.md (sources/pricing.md L127 (under "Is OneDroid Synapse free?"))
    The free hosted tier including the event log is stated on the pricing and Synapse pages; no public terms or billing artefact backs it.
    C-522including the event log, which is not a paid tieronedroid.ai/llms-full.txt (sources/llms-full.txt L4172 (PAGE /synapse; under "Is OneDroid Synapse free?"))That the event log is not a paid tier is a pricing statement repeated across pages, with nothing beyond the statement.
    C-523tool call is authenticated, policy-checked and logged while the hub's activity logging is on (fail-open: a logging failure never blocks a call), to an audit log you ownonedroid.ai/llms-full.txt (sources/llms-full.txt L4398 (PAGE /quickstart; under "Connect an agent to one governed URL."))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L4974-4975 (PAGE /financial-services; under "Evidence as a by-product"))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L5459-5461 (PAGE /kit; under "Eyes and hands: OneDroid Synapse"))
    The same Synapse per-call authentication, policy check and audit log statement, restated with no public code or sample log.
    C-525the scenarios are written once per feature, before the build, and run as often as you needonedroid.ai/llms-full.txt (sources/llms-full.txt L5260-5261 (PAGE /qa-teams; under "What may differ"))A process description of how scenarios are written and run, with no artefact beyond the statement.
    C-527Give this page to your coding agent and it sets you up, stopping where it needs your decision.onedroid.ai/llms-full.txt (sources/llms-full.txt L5394 (PAGE /kit))The runbook does mark HUMAN stops, but no public record shows an agent completing the setup from the page; a published validation report from such a run would be needed.
    C-537If the report is not in the repo, the setup is not done, whatever the agent says.onedroid.ai/llms-full.txt (sources/llms-full.txt L4579-4580 (PAGE /quickstart; under "The check"))A process rule; the runbook repeats that validation commits its report to the human's repo, but that is the same rule restated.
    C-541No script signs in for you or grants access on your behalf.onedroid.ai/llms-full.txt (sources/llms-full.txt L5505-5506 (PAGE /kit; under "The method: the Dark Factory kits"))The runbook repeats that no script can sign in or grant access, but proving no script does it would take an audit of every kit script.
    C-545Never report a step as done without its check.onedroid.ai/llms-full.txt (sources/llms-full.txt L5514-5515 (PAGE /kit; under "The steps, in the same order, each with its check."))A conduct instruction to the agent, repeated in the runbook, with nothing that enforces it.
    C-551The hub from step 03 is the one to record here: it gives the agent memory across sessions and connectors to the systems the human already uses. The method also runs without one.onedroid.ai/llms-full.txt (sources/llms-full.txt L5661-5663 (PAGE /kit; under "Connect the human's tools"))The runbook repeats that the hub gives memory and connectors and that the method runs without one; nothing beyond the statement shows it.
    C-553Then open a new session, because tools and hooks load only when a session starts.onedroid.ai/llms-full.txt (sources/llms-full.txt L5663-5664 (PAGE /kit; under "Connect the human's tools"))A statement about the agent harness's loading behaviour, with no publisher artefact behind it.
    C-557You skip the lockfile, the pins and the validation that way.onedroid.ai/llms-full.txt (sources/llms-full.txt L5691-5692 (PAGE /kit; under "Only the skills?"))The skills index answers 200, but that taking only the skills skips the lockfile, pins and validation is a consequence stated on /kit with no artefact behind it.
    C-559Quickstart: a OneDroid Synapse hub and OneDroid Engram memory in about ten minutes eachonedroid.ai/llms-full.txt (sources/llms-full.txt L5719-5720 (PAGE /kit; under "2 and 3 · Synapse and Engram"))The ten-minutes-each figure is repeated on /kit and llms.txt; timed walkthrough records would be needed.
    C-562the Argus control plane, hosted by default, moves into your environment on requestonedroid.ai/llms-full.txt (sources/llms-full.txt L5847 (PAGE /enterprise; under "OneDroid Argus, Synapse and Engram, on your infrastructure."))The repo confirms the control plane is a hosted service not in the repo, but moving it into the customer's environment on request is stated only.
    C-570For partners and contacts: this page is written to be forwarded.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L257 (under "Your architecture says what may talk to what. Does the running system agree?"))A statement of the page's intent; nothing beyond the page backs it.
    C-572FINOS states the boundary of that file itself, in its own controls tutorial: "Validation checks the documented configuration, not whether the running system meets the requirement."onedroid.ai/partners-financial-services (sources/partners-financial-services.html L279-281 (under "01An architecture file is a list of claims"))The FINOS controls tutorial is linked but is not among the pages fetched in D/fetched/, so the quote is unchecked here and rests on the page.
    C-573CALM declares. OneDroid Argus measures. The two are complementary.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L288 (under "01An architecture file is a list of claims"))A positioning line; nothing beyond the page backs it.
    C-583There is no number to optimise for and no "safe" label.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L323 (under "02What OneDroid Argus does"))The anchored verdict payload has a verdict and tallies and no score field, but that absence in one record cannot show the product never offers a number or label.
    C-586The one hand-written check, a second way to the same trade, is the one that found it.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L351-352 (under "03The proof: a FINOS CALM trading demo"))The chain tallies show 2 failures but not which checks; the per-check results sit in a private bundle known only by hash.
    C-587The cause was ours: the cluster we ran on does not enforce Kubernetes network policies.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L354-355 (under "03The proof: a FINOS CALM trading demo"))The cause is described on the pages; the cluster's network-policy configuration and the run reports are not published.
    C-591Two certified runs on one cluster the same day, in-memory demo data, no capacity claim.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L365 (under "03The proof: a FINOS CALM trading demo"))The limits are stated; the earlier certified run's records and the demo's data handling are not published beyond the private bundle hash.
    C-592The same team wrote the checks and ran them: the seal proves the checks existed before the result, not that their authors were independent.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L365-367 (under "03The proof: a FINOS CALM trading demo"))A disclosure by the publisher; nothing beyond the statement shows who wrote and ran the checks.
    C-593Nothing was reported to the demo's authors.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L367 (under "03The proof: a FINOS CALM trading demo"))A negative statement repeated in the case study; no artefact can show the absence of a report.
    C-594FINOS has published a set of SDLC Common Controls as drafts.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L377 (under "04Where it fits next to the FINOS SDLC Common Controls"))The FINOS controls site is linked but not in D/fetched/, so their draft status is unchecked here.
    C-595Most of them are evidenced from the repository and the pipeline.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L377-378 (under "04Where it fits next to the FINOS SDLC Common Controls"))The publisher's reading of the controls; a per-control mapping against the FINOS text would be needed.
    C-596A few depend on what the running system does, and that is the part OneDroid Argus reaches.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L378-379 (under "04Where it fits next to the FINOS SDLC Common Controls"))The publisher's reading of which controls depend on runtime; nothing beyond the page backs it.
    C-597This is our reading of the controls, not FINOS's.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L379 (under "04Where it fits next to the FINOS SDLC Common Controls"))A self-disclosure; nothing beyond the statement.
    C-602Not reached: OneDroid Argus never reads the diffonedroid.ai/partners-financial-services (sources/partners-financial-services.html L389 (under "04Where it fits next to the FINOS SDLC Common Controls"))The runner's documented inputs are scenarios, config and the running system, but no artefact can show it never reads a diff.
    C-603For years companies spent between 23% and 35% of their IT budget on quality assurance and testing, according to Capgemini's World Quality Report, and testing was the part that got cut.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L398-400 (under "05What it costs"))The yearly figures are restated on /argus; the Capgemini World Quality Report is not in D/fetched/, so the range is unchecked here.
    C-604In one measured experiment, Claude Code built a feature for $1.01 to $1.97 a run, and OneDroid Argus as the independent tester cost one session of about $5.30 for the whole feature.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L402-404 (under "05What it costs"))The case study's means are internally consistent (5.30/3 = 1.77, 1.01 + 1.77 = 2.78), but no per-run cost logs are published, and $1.01 and $1.97 are the two setups' means, not a range for one setup.
    C-605Its checks stay as regression checks for the next change.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L404 (under "05What it costs"))Reuse of checks as regression checks is stated; a published reused set across changes would be needed.
    C-608Not a code reviewer. It never reads the diff.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L420 (under "06What it is not"))A negative capability; the documented inputs exclude the diff, but nothing can show it is never read.
    C-610Not proof the builder never saw the tests. That is custody: a credential that cannot read a test, and an environment that does not hold one. The ledger proves priority, not that the builder never looked.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L424-426 (under "06What it is not"))A stated limit of the ledger; nothing beyond the statement.
    C-611Not a judgement of the architecture. A run can say whether the running system keeps what the architecture declares, not whether the architecture is a good one.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L427-428 (under "06What it is not"))A stated limit repeated in the case study; nothing beyond the statement.
    C-614You reach Michal Bacia and the team that builds the product, not a sales queue.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L443-444 (under "07How to start"))A mailto link to michal@onedroid.ai is present; who answers is stated only.
    C-615OneDroid Argus for financial services, October 2026. Figures are from the pages they link to, read 23 October 2026.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L448-449 (under "07How to start"))The page's own date stamp; nothing beyond the statement shows the read date.
    C-616FINOS and CALM are names that belong to their owners; OneDroid is not affiliated with FINOS or the CALM project.onedroid.ai/partners-financial-services (sources/partners-financial-services.html L449-450 (under "07How to start"))A trademark and non-affiliation disclaimer; nothing beyond the statement.
    C-623These are the two ways to start Argus; the site's three uses (holdout acceptance, under load, continuous) all run through them.docs.onedroid.ai/argus (sources/argus-docs.html L149 (under "The two ways to start a run"))
    docs.onedroid.ai/argus.md (sources/argus-docs.md L95 (under "The two ways to start a run"))
    The run/monitor split and the mapping of the three uses are documentation; scheduling lives in the closed control plane.
    C-626In one measured experiment, Claude Code built a feature for $1.01 to $1.97 a run, and the independent tester's session, which wrote the checks once for the whole feature, cost about $5.30onedroid.ai/llms-full.txt (sources/llms-full.txt L2802-2804 (PAGE /argus; under "Testing was the part teams could skip. Not any more."))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L5439-5441 (PAGE /kit; under "Testing: OneDroid Argus"))
    onedroid.ai/argus.md (sources/argus.md L66-68 (under "Testing was the part teams could skip. Not any more."))
    Means in the case-study table only, internally consistent, but no per-run cost or token logs are published.
    C-627It is the evaluation environment: for testing, with no promise that data is kept, so keep production data out of it.onedroid.ai/llms-full.txt (sources/llms-full.txt L2942-2943 (PAGE /argus; under "MCP server"))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L5526-5527 (PAGE /kit; under "Get an account and the CLI"))
    onedroid.ai/argus.md (sources/argus.md L206-207 (under "MCP server"))
    The evaluation-environment caveat is repeated in the repo README and on the pages; no published data-retention terms for argus-dev back it.
    C-629In a dark software factory, AI agents do the work, mission in and verified software out, and evidence, not the agents' word, decides what ships.onedroid.ai/llms-full.txt (sources/llms-full.txt L5403-5404 (PAGE /kit; under "The kit to build your"))A definition of the term; nothing beyond the statement.
    C-630It takes four parts.onedroid.ai/llms-full.txt (sources/llms-full.txt L5404 (PAGE /kit; under "The kit to build your"))The page lists four parts; the count is self-referential to the page.
    C-631An agent that tests its own work tests its own reading of the ticket, and a model that can see its checker learns to satisfy the checker.onedroid.ai/llms-full.txt (sources/llms-full.txt L5428-5429 (PAGE /kit; under "Testing: OneDroid Argus"))An argument about self-testing with no study or data on the publisher's surfaces behind it here.
    C-634The checks stay as regression checks for the next change.onedroid.ai/llms-full.txt (sources/llms-full.txt L5436-5437 (PAGE /kit; under "Testing: OneDroid Argus"))Reuse as regression checks is stated; a reused set across changes would be needed.
    C-638Access exists only for accounts you connect, limited to the permissions you grantonedroid.ai/llms-full.txt (sources/llms-full.txt L5458-5459 (PAGE /kit; under "Eyes and hands: OneDroid Synapse"))A per-tool toggle exists in the app, but that access is limited to connected accounts and granted permissions is not shown without the closed gateway code.
    C-639Versioned, permissioned memory in a Postgres you control with your own database, or one OneDroid hosts, searched by vector, keyword and graph at once, and reachable by any MCP client behind any model.onedroid.ai/llms-full.txt (sources/llms-full.txt L5474-5476 (PAGE /kit; under "Memory and context: OneDroid Engram"))Engram's code is not public here; a schema, BYODB setup doc or search API sample would be needed.
    C-640What one agent wrote, the next one can find.onedroid.ai/llms-full.txt (sources/llms-full.txt L5476-5477 (PAGE /kit; under "Memory and context: OneDroid Engram"))A capability statement with no public artefact.
    C-641Every write is a commit: diff any two versions, or restore a previous one in one call.onedroid.ai/llms-full.txt (sources/llms-full.txt L5477-5478 (PAGE /kit; under "Memory and context: OneDroid Engram"))Diff and restore are stated; a public API reference or tool listing would be needed.
    C-642Included with a Synapse account, and it runs on its own too.onedroid.ai/llms-full.txt (sources/llms-full.txt L5480-5481 (PAGE /kit; under "Memory and context: OneDroid Engram"))Bundling and standalone operation are stated on /kit and /pricing only.
    C-643Parts 1 to 3 are tools.onedroid.ai/llms-full.txt (sources/llms-full.txt L5488 (PAGE /kit; under "The method: the Dark Factory kits"))A framing of the page's own structure; nothing beyond the statement.
    C-645Agent: do the parts in order.onedroid.ai/llms-full.txt (sources/llms-full.txt L5512 (PAGE /kit; under "The steps, in the same order, each with its check."))An instruction to the reading agent on the /kit page; nothing beyond the page itself can back an ordering instruction.
    C-646Where a step links to a runbook, the runbook wins over this summary if the two ever disagree.onedroid.ai/llms-full.txt (sources/llms-full.txt L5515-5516 (PAGE /kit; under "The steps, in the same order, each with its check."))A precedence rule stated on the /kit page only; no artefact shows a runbook-over-summary rule being applied.
    C-649The session that writes the checks is never the session that writes the code.onedroid.ai/llms-full.txt (sources/llms-full.txt L5543 (PAGE /kit; under "Keep the tester and the builder apart"))role.go splits product and test hats by token, but that two different sessions write the checks and the code is a working practice no artefact shows.
    C-652The token is shown once.onedroid.ai/llms-full.txt (sources/llms-full.txt L5559 (PAGE /kit; under "Create a hub and connect the agent"))Synapse answers 200 but token creation sits behind sign-in, which was not done, so 'shown once' is stated only.
    C-653Check: `claude mcp list` shows `synapse … ✓ Connected`.onedroid.ai/llms-full.txt (sources/llms-full.txt L5568 (PAGE /kit; under "Create a hub and connect the agent"))The expected 'claude mcp list' output is quoted on /kit and /quickstart; it needs a token and a client session to observe, so nothing beyond the statement backs it.
    C-654Enabling a service on the hub and saving the human's login for it are two different actions, and only the second lets the agent call anything: if the agent connects and sees no tools, this is why.onedroid.ai/llms-full.txt (sources/llms-full.txt L5569-5571 (PAGE /kit; under "Create a hub and connect the agent"))The enable-versus-save-login distinction is stated on /kit and /quickstart; the Synapse code is not public and the behaviour sits behind sign-in.
    C-656Check: the write returns an object id, and the search returns that object.onedroid.ai/llms-full.txt (sources/llms-full.txt L5586 (PAGE /kit; under "Write one thing, search it back"))The write-returns-id and search-returns-object behaviour needs a token to exercise; the endpoint refuses an unauthenticated call, so the result is stated only.
    C-658That is why the kit starts there.onedroid.ai/llms-full.txt (sources/llms-full.txt L5742-5743 (PAGE /kit; under "Testing first. Always."))The kit page does put Testing as Part 1, but the 'that is why' rationale is framing with nothing beyond the page behind it.
    C-659OpenAI computes the search embeddings before they are storedonedroid.ai/llms-full.txt (sources/llms-full.txt L5964 (PAGE /pricing; under "Hosted, any team size"))
    onedroid.ai/pricing.md (sources/pricing.md L79 (under "Hosted, any team size"))
    OpenAI as the embedding provider is stated on /pricing and /privacy; Engram code and provider configuration are not public, so nothing beyond the statement backs it.
    C-660We test and audit agents in situ: what they did, what they claimed, what it cost, which human saw which decision, and what changes when one parameter changes.onedroid.ai/llms-full.txt (sources/llms-full.txt L28-29 (PAGE /; under "Test the agents where they work: in the workflow, in the fleet, next to the people who sign off."))The in-situ testing and audit scope for agents is stated on the home page, llms.txt and /argus-agents, which itself marks the agent product in development; no artefact shows agent audits.
    C-661OneDroid Argus tests whether that is still true once the agents move in: are the humans in the loop deciding, or rubber-stamping?onedroid.ai/llms-full.txt (sources/llms-full.txt L33-34 (PAGE /; under "Test the agents where they work: in the workflow, in the fleet, next to the people who sign off."))The rubber-stamping test is stated on the home page and llms.txt; repo-grep found nothing and no workflow scenario is published.
    C-663In development: Argus for agentic workflows and fleets, in two shapes: one free, open-source binary on a developer's machine, and the record and the tests on an enterprise's own estate.onedroid.ai/llms-full.txt (sources/llms-full.txt L37-39 (PAGE /; under "Test the agents where they work: in the workflow, in the fleet, next to the people who sign off."))An in-development status repeated on home, llms.txt, /pricing and /argus-agents; no repo or release for either shape exists to check.
    C-670We found no study that tested the promise on a running agent.onedroid.ai/llms-full.txt (sources/llms-full.txt L73 (PAGE /; under "Why we test in situ, and not in a lab."))A negative literature-search claim stated on the home page with no published search record, query list or bibliography behind it.
    C-672A long-lived scientific program nobody on our team wrote, tested three ways without reading its source first: against its manual, against an earlier release, and an AI rewrite against the original.onedroid.ai/llms-full.txt (sources/llms-full.txt L89 (PAGE /; under "Three ways to test code you did not write"))
    onedroid.ai/llms-full.txt (sources/llms-full.txt L3133 (PAGE /case-studies; under "Three ways to test code you did not write"))
    The case study's summary is repeated on home, /case-studies and the study page; the program is unnamed and the run files are private, so nothing beyond the statement backs it.
    C-674the scenarios are written once per feature, before the build, and stay as regression checks for the next changeonedroid.ai/llms-full.txt (sources/llms-full.txt L287 (PAGE /; under "Is OneDroid Argus end-to-end testing?"))
    onedroid.ai/llms.txt (sources/llms.txt L42 (under "OneDroid"))
    Write-once-per-feature and regression reuse are stated on home, llms.txt and a case study ('ran 37 older checks'); no published scenario set or run log shows the reuse.
    C-675This page is Argus Code, the part of Argus that tests code. Argus for agentic workflows and fleets, with a free local kit for developers, is in developmentonedroid.ai/llms-full.txt (sources/llms-full.txt L2765-2766 (PAGE /argus; under "Independent testing, at the same order of cost as the coding."))
    onedroid.ai/argus.md (sources/argus.md L29-30 (under "Independent testing, at the same order of cost as the coding."))
    A naming and status statement on /argus; the in-development kit has no repo or release to check.
    C-677That part is now called Argus Code.onedroid.ai/llms-full.txt (sources/llms-full.txt L3015-3016 (PAGE /argus-agents; under "Test the agents where they run,"))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L16-17 (under "Test the agents where they run,"))
    A naming statement; the site uses 'Argus Code' across pages, but the runner README does not, and naming has no artefact beyond the pages.
    C-678The next part tests and audits the agents themselves, where they run: single coding agents on a developer's machine, and agentic workflows and fleets on a company's own estate.onedroid.ai/llms-full.txt (sources/llms-full.txt L3016-3017 (PAGE /argus-agents; under "Test the agents where they run,"))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L17-18 (under "Test the agents where they run,"))
    A description of an in-development product; no repo, release or record exists to check.
    C-682It records every session of your coding agent (prompts, tool calls, results, model usage), writes the cost per model into each commit and sums it per branch and ticket, checks prompts and tool calls for credentials as they happenonedroid.ai/llms-full.txt (sources/llms-full.txt L3046-3048 (PAGE /argus-agents; under "One binary, on your machine, nothing leaves it."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L47-49 (under "One binary, on your machine, nothing leaves it."))
    Capabilities of a binary the same page marks in development; no repo or release exists to check.
    C-683at the end of a session tells you what the agent actually did against what it said it did, with the receipt: files touched, commands and exit codes, tests that ran, test files edited, tools used, the diff, and beside every green line what the hooks could not see.onedroid.ai/llms-full.txt (sources/llms-full.txt L3048-3051 (PAGE /argus-agents; under "One binary, on your machine, nothing leaves it."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L49-52 (under "One binary, on your machine, nothing leaves it."))
    The session verdict is listed as in development on the same page (L61); nothing beyond the description exists.
    C-684The same binary runs a set of open basic checks against your agent, locally, with a pass list you can verify without an account.onedroid.ai/llms-full.txt (sources/llms-full.txt L3051-3052 (PAGE /argus-agents; under "One binary, on your machine, nothing leaves it."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L52-53 (under "One binary, on your machine, nothing leaves it."))
    The open check set is listed as in development on the same page; no published checks or pass list.
    C-685Claude Code first, then Codex CLI and Gemini CLI.onedroid.ai/llms-full.txt (sources/llms-full.txt L3054 (PAGE /argus-agents; under "One binary, on your machine, nothing leaves it."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L55 (under "One binary, on your machine, nothing leaves it."))
    A support-order roadmap for an unreleased binary; nothing to check.
    C-686Open source under Apache 2.0, the basic checks included, with contributions by sign-off.onedroid.ai/llms-full.txt (sources/llms-full.txt L3054-3055 (PAGE /argus-agents; under "One binary, on your machine, nothing leaves it."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L55-56 (under "One binary, on your machine, nothing leaves it."))
    An Apache 2.0 licence for a binary not yet released; the only open repository is MIT (only its jmeter AMQP plugin is Apache), so no Apache-2.0 artefact for this product exists.
    C-687Every open check is labelled a smoke test, not a certificateonedroid.ai/llms-full.txt (sources/llms-full.txt L3055-3056 (PAGE /argus-agents; under "One binary, on your machine, nothing leaves it."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L56-57 (under "One binary, on your machine, nothing leaves it."))
    A labelling commitment for an open check set that is not yet published.
    C-689It will be announced here and on the changelog when it can be installed.onedroid.ai/llms-full.txt (sources/llms-full.txt L3061 (PAGE /argus-agents; under "One binary, on your machine, nothing leaves it."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L62 (under "One binary, on your machine, nothing leaves it."))
    A future announcement commitment; nothing can back it before it happens.
    C-690A recorder on every machine and agent runtime you run, feeding a store you own.onedroid.ai/llms-full.txt (sources/llms-full.txt L3067 (PAGE /argus-agents; under "Every agent recorded, every workflow tested, nothing leaves your estate."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L68 (under "Every agent recorded, every workflow tested, nothing leaves your estate."))
    The central store is listed as in development on the same page (L86-88); no artefact.
    C-691The full record of each session written to a private ledger you run, so that an agent cannot quietly rewrite its own history.onedroid.ai/llms-full.txt (sources/llms-full.txt L3069-3070 (PAGE /argus-agents; under "Every agent recorded, every workflow tested, nothing leaves your estate."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L70-71 (under "Every agent recorded, every workflow tested, nothing leaves your estate."))
    The private session ledger for agents is part of an in-development product; the ledger library named at L85 is not public.
    C-692One dashboard: which agents are alive, recording or silent; spend by team with budgets; anomalies against your own baselines; alerts under rules a person sets (token burn, rewrites, velocity, rejected submissions) with the time each took to reach a person; a pause button for any agent, the pause itself on the record; a queue where cases carry a label and only a named person moves them.onedroid.ai/llms-full.txt (sources/llms-full.txt L3070-3074 (PAGE /argus-agents; under "Every agent recorded, every workflow tested, nothing leaves your estate."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L71-75 (under "Every agent recorded, every workflow tested, nothing leaves your estate."))
    The dashboards and fleet alerts are listed as in development on the same page; no artefact.
    C-693Tests run in situ: the runner on your estate, triggered from the Argus control plane, which receives verdicts, hashes and alerts and never the content.onedroid.ai/llms-full.txt (sources/llms-full.txt L3076-3077 (PAGE /argus-agents; under "Every agent recorded, every workflow tested, nothing leaves your estate."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L77-78 (under "Every agent recorded, every workflow tested, nothing leaves your estate."))
    The runner exists, but in-situ workflow tests for agents and a control plane that never receives content are not shown by any artefact.
    C-694Workflow scenarios take several agents as the subject, replay a recorded run with one parameter changed, and ask whether the humans in the loop are doing the decidingonedroid.ai/llms-full.txt (sources/llms-full.txt L3077-3079 (PAGE /argus-agents; under "Every agent recorded, every workflow tested, nothing leaves your estate."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L78-80 (under "Every agent recorded, every workflow tested, nothing leaves your estate."))
    Workflow scenarios and replay are listed as in development on the same page; no artefact.
    C-696The first deployment is a lending workflow for a client, with its lender as the reader of the record. Pricing is by contact until that deployment is delivered.onedroid.ai/llms-full.txt (sources/llms-full.txt L3087-3088 (PAGE /argus-agents; under "Every agent recorded, every workflow tested, nothing leaves your estate."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L88-89 (under "Every agent recorded, every workflow tested, nothing leaves your estate."))
    An unnamed client deployment and a pricing policy, stated only.
    C-697nothing we ship offers to replace a person's judgement, test or decision, and nothing shifts responsibility to an AIonedroid.ai/llms-full.txt (sources/llms-full.txt L3098-3099 (PAGE /argus-agents; under "The human stays in the driving seat."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L99-100 (under "The human stays in the driving seat."))
    A conduct rule stated on /argus-agents; no artefact can show the absence of judgement-replacing features across everything shipped.
    C-698everything comes with a full record, replay, and a kill switch a person can pull at once, with the stop itself on the recordonedroid.ai/llms-full.txt (sources/llms-full.txt L3099-3101 (PAGE /argus-agents; under "The human stays in the driving seat."))
    onedroid.ai/argus-agents.md (sources/argus-agents.md L100-102 (under "The human stays in the driving seat."))
    The runner has a replay command, but 'everything comes with' a full record, replay and a recorded kill switch is a blanket claim no artefact covers.
    C-700The files behind this page (the certificate and run reports of both certified runs, the seven checks and the load checks, the load-ramp record, the refused run, the demo's manifests and network policies, the verify output) are held by OneDroid and shown on request, so that a reader or a claims register can check them: `private: OneDroid, shared on request | sha256:37e5df715c2f8ee3670704caebfb06a0b7fb17883450b02ee31fa3284e49a1b0`.onedroid.ai/llms-full.txt (sources/llms-full.txt L3514 (PAGE /case-study-finos-calm; under "How anyone can check it later"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L141 (under "How anyone can check it later"))
    The files are private and the sha256 is of a bundle shown only on request, so neither the hash nor the contents can be checked from public surfaces.
    C-701The files behind every number on this page (the per-run tokens and costs as recorded, the 13 judge checks, the cost cross-check, the review) are held by OneDroid and shown on request, so that a reader or a claims register can check them: `private: OneDroid, shared on request | sha256:759cf275abe53c1dfbf3d72efb7e8049dc67ca1593c0644e9f4b32f5ca15aeb1`.onedroid.ai/llms-full.txt (sources/llms-full.txt L3678 (PAGE /case-study-claude-code-alone-vs-tester; under "What this experiment did not cover"))
    onedroid.ai/case-study-claude-code-alone-vs-tester.md (sources/case-study-claude-code-alone-vs-tester.md L115 (under "What this experiment did not cover"))
    A private bundle with a quoted sha256; nothing public to hash or read.
    C-702The files behind every number on this page (the three ports per setup, the 215 equivalence checks and their hidden and visible halves, the per-run judge outputs, the costs as recorded, the review) are held by OneDroid and shown on request, so that a reader or a claims register can check them: `private: OneDroid, shared on request | sha256:d454c4975cd1fbf90ebf353a0af6b18fde2926e2bf47abc19a0b54601f20e467`.onedroid.ai/llms-full.txt (sources/llms-full.txt L3802 (PAGE /case-study-modernization-three-setups; under "What this experiment did not cover"))
    onedroid.ai/case-study-modernization-three-setups.md (sources/case-study-modernization-three-setups.md L111 (under "What this experiment did not cover"))
    A private bundle with a quoted sha256; nothing public to hash or read.
    C-703Case study. An experiment by our team. Every run had a written prediction beforehand.onedroid.ai/llms-full.txt (sources/llms-full.txt L3831 (PAGE /case-study-three-ways-to-test-code; under "Three ways to test code you did not write"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L16 (under "Three ways to test code you did not write"))
    The predictions are not published, so 'every run had a written prediction' is stated only.
    C-704We tested a long-lived scientific program nobody on our team wrote, without reading its source first, by comparing it with three things: its manual, an earlier release and the original program.onedroid.ai/llms-full.txt (sources/llms-full.txt L3835 (PAGE /case-study-three-ways-to-test-code; under "Three ways to test code you did not write"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L20 (under "Three ways to test code you did not write"))
    The method summary of an unnamed program with private run files; nothing beyond the statement.
    C-705The manual showed where code and documentation disagree, including some crashes. The earlier release showed no regression, and one old run that exited cleanly with a badly wrong number.onedroid.ai/llms-full.txt (sources/llms-full.txt L3836 (PAGE /case-study-three-ways-to-test-code; under "Three ways to test code you did not write"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L21 (under "Three ways to test code you did not write"))
    Results with no published checks, reports or counts; stated only.
    C-707OneDroid Argus doesn't need to know what the code is meant to do.onedroid.ai/llms-full.txt (sources/llms-full.txt L3846 (PAGE /case-study-three-ways-to-test-code; under "In short"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L31 (under "In short"))
    A capability framing; the runner compares systems, but no artefact shows a test needing no intent, and the study's files are private.
    C-708Against an earlier release: no regression over these checks, and one old run that exited cleanly with a plausible file and a main number well off.onedroid.ai/llms-full.txt (sources/llms-full.txt L3850 (PAGE /case-study-three-ways-to-test-code; under "In short"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L35 (under "In short"))
    A regression result with no published checks or run output; stated only.
    C-709Agents that read only the manual wrote requirements, each quoting the manual with a page number, and a script checked that every quote was on that page.onedroid.ai/llms-full.txt (sources/llms-full.txt L3856 (PAGE /case-study-three-ways-to-test-code; under "1. Does the code do what its manual says?"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L41 (under "1. Does the code do what its manual says?"))
    The requirement quotes and the page-check script are not published; stated only.
    C-710A person read each red against the source, and our own mistakes were fixed and not counted.onedroid.ai/llms-full.txt (sources/llms-full.txt L3858 (PAGE /case-study-three-ways-to-test-code; under "1. Does the code do what its manual says?"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L43 (under "1. Does the code do what its manual says?"))
    Human review of reds and excluded mistakes are not recorded anywhere public.
    C-711Result: places where the code and its manual disagree. Most mean the manual is out of date. The rest are wrong behaviour, and some of those end in a crash or a runtime error.onedroid.ai/llms-full.txt (sources/llms-full.txt L3860 (PAGE /case-study-three-ways-to-test-code; under "1. Does the code do what its manual says?"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L45 (under "1. Does the code do what its manual says?"))
    Disagreement classes with no counts or published reds; stated only.
    C-712The same checks ran against the current release and an older one, side by side.onedroid.ai/llms-full.txt (sources/llms-full.txt L3864 (PAGE /case-study-three-ways-to-test-code; under "2. Did anything change between two releases?"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L49 (under "2. Did anything change between two releases?"))
    No run record for the side-by-side compare is published (section 2 is not even in the private bundles listed at L83).
    C-713The checks that went red on the old release only had each been predicted from the project's own release notes and reference files.onedroid.ai/llms-full.txt (sources/llms-full.txt L3866 (PAGE /case-study-three-ways-to-test-code; under "2. Did anything change between two releases?"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L51 (under "2. Did anything change between two releases?"))
    The predictions and the red list are not published.
    C-714On one input, the old release exited cleanly, reported no error and wrote a plausible file, and its main number was well off.onedroid.ai/llms-full.txt (sources/llms-full.txt L3868 (PAGE /case-study-three-ways-to-test-code; under "2. Did anything change between two releases?"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L53 (under "2. Did anything change between two releases?"))
    One old-release result with no input, output or number published.
    C-716Result: several kinds of difference, most of them wrong numbers with a clean exit. Some totals were many times too high, and some values came out negative where that makes no sense.onedroid.ai/llms-full.txt (sources/llms-full.txt L3874 (PAGE /case-study-three-ways-to-test-code; under "3. Does an AI rewrite do what the original does?"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L59 (under "3. Does an AI rewrite do what the original does?"))
    Difference kinds with no published values or judge outputs.
    C-719The files behind sections 1 and 3 are held by OneDroid and shown on request, so that a reader or a claims register can check them. Section 1, the manual: the certificate, the final and rehearsal reports, the sealed checks, the predictions, the build record and the second-model review: `private: OneDroid, shared on request | sha256:a508ecf47d33df4b3cbceec6e194f59e54ac8337255b41073e297db57e56921c`. Section 3, the rewrite: the ports, the checks and their split, the judge outputs and the costs as recorded: `private: OneDroid, shared on request | sha256:d454c4975cd1fbf90ebf353a0af6b18fde2926e2bf47abc19a0b54601f20e467`.onedroid.ai/llms-full.txt (sources/llms-full.txt L3898 (PAGE /case-study-three-ways-to-test-code; under "What this does not show"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L83 (under "What this does not show"))
    Two private bundles with quoted sha256 values; repo-grep found no 'second-model review', and nothing public to hash.
    C-720The program, its authors and its owners are not named, and the details that would identify them are left out on purpose.onedroid.ai/llms-full.txt (sources/llms-full.txt L3900 (PAGE /case-study-three-ways-to-test-code; under "What this does not show"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L85 (under "What this does not show"))
    The saved page names no program, author or owner, which fits the claim, but the deliberate omission is a statement of intent only.
    C-721On the hosted gateway, OneDroid holds the encrypted credentials and the event log (see /security).onedroid.ai/llms-full.txt (sources/llms-full.txt L4127-4128 (PAGE /synapse; under "Ownership"))/security repeats that the hosted gateway holds encrypted credentials and the event log; the Synapse code is not public, so it is restated.
    C-722The next part of Argus records every agent you run, into a store you own, with the full record of each session on a private ledger you run; tests whole workflows in situ, with the runner on your estate triggered from the control plane, which receives verdicts, hashes and alerts and never the content; and puts the fleet on one dashboard with alerts under rules a person sets, a pause for any agent and a queue only a named person moves. Nothing leaves your estate. It is in development and priced by contactonedroid.ai/llms-full.txt (sources/llms-full.txt L5871-5875 (PAGE /enterprise; under "In development: Argus for agentic workflows and fleets, on your estate"))The /enterprise summary of the in-development fleet product; same statement as /argus-agents, no artefact.
    C-723the Argus kit for developers that is in development will be free and open sourceonedroid.ai/llms-full.txt (sources/llms-full.txt L5900-5901 (PAGE /pricing; under "Free to use."))
    onedroid.ai/pricing.md (sources/pricing.md L15-16 (under "Free to use."))
    A future licence and price commitment for an unreleased kit.
    C-724Enterprises get the products on their own infrastructure, with org-wide governance, priced by contact.onedroid.ai/llms-full.txt (sources/llms-full.txt L5902 (PAGE /pricing; under "Free to use."))
    onedroid.ai/pricing.md (sources/pricing.md L17 (under "Free to use."))
    On-premises enterprise deployment with org-wide governance is listed on /pricing; repo-grep found nothing and no enterprise package is public.
    C-725In development, two more shapes of Argus. For developers: one free, open-source binary that records your coding agent's sessions, gives the verdict on what it did, counts the cost and runs the open basic checks, all on your machine. For enterprises: the record and the tests on your own estate, priced by contact. Neither can be installed yetonedroid.ai/llms-full.txt (sources/llms-full.txt L5946-5949 (PAGE /pricing; under "The OneDroid Argus control plane"))
    onedroid.ai/pricing.md (sources/pricing.md L61-64 (under "The OneDroid Argus control plane"))
    The in-development status and 'neither can be installed' are restated across /pricing and /argus-agents; no artefact.
    C-726Annual licence, priced per deployment. Email with your shape (cloud or on-premises, how many hubs and agents) and you get a number back.onedroid.ai/llms-full.txt (sources/llms-full.txt L5992-5993 (PAGE /pricing; under "Enterprise"))
    onedroid.ai/pricing.md (sources/pricing.md L107-108 (under "Enterprise"))
    A licensing and pricing model stated on /pricing; no licence terms or price sheet is published.
    C-727It is an annual licence, priced per deployment. Contact OneDroid with your deployment shape for a number; there is no public price list for the enterprise tier.onedroid.ai/llms-full.txt (sources/llms-full.txt L6020 (PAGE /pricing; under "How much does OneDroid Enterprise cost?"))
    onedroid.ai/pricing.md (sources/pricing.md L135 (under "How much does OneDroid Enterprise cost?"))
    The same per-deployment licence, restated in the FAQ; 'no public price list' fits the pages but is a statement.

    Evidence available on request (27)

    IdClaim, as madeMade atWhat we were shown
    C-010It used 0.72 million tokens against 2.37 milliononedroid.ai/llms-full.txt (sources/llms-full.txt L3598 (PAGE /case-study-claude-code-alone-vs-tester; under "In short"))
    onedroid.ai/case-study-claude-code-alone-vs-tester.md (sources/case-study-claude-code-alone-vs-tester.md L35 (under "In short"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 759cf275abe53c1dfbf3d72efb7e8049dc67ca1593c0644e9f4b32f5ca15aeb1
    Per-run token table; mean tokens_total of B-1..B-3 = 0.72 million and of A-1..A-3 = 2.37 million.
    C-011It is split evenly over the three B runs, which adds about $1.77 to each.onedroid.ai/llms-full.txt (sources/llms-full.txt L3652 (PAGE /case-study-claude-code-alone-vs-tester; under "The result"))
    onedroid.ai/case-study-claude-code-alone-vs-tester.md (sources/case-study-claude-code-alone-vs-tester.md L89 (under "The result"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 759cf275abe53c1dfbf3d72efb7e8049dc67ca1593c0644e9f4b32f5ca15aeb1
    Per-run cost table; tester_usd is the same $1.77 on each of B-1..B-3, three shares summing to about $5.30.
    C-013Every port was judged on the same 109 hidden checks against the original's answers: A passed 104, B 81, B′ 106.onedroid.ai/llms-full.txt (sources/llms-full.txt L3726 (PAGE /case-study-modernization-three-setups; under "In short"))
    onedroid.ai/case-study-modernization-three-setups.md (sources/case-study-modernization-three-setups.md L35 (under "In short"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 d454c4975cd1fbf90ebf353a0af6b18fde2926e2bf47abc19a0b54601f20e467
    Per-run judge outputs on all 215 checks, restricted to the frozen 109-name JUDGE list, recompute to 104 for every A run, a B mean of 81 and 106 for every B′ run, matching the judge_half_passed column of the per-run table.
    C-014B′ cost $10.04 a run, 41% less than A at $16.92.onedroid.ai/llms-full.txt (sources/llms-full.txt L3728 (PAGE /case-study-modernization-three-setups; under "In short"))
    onedroid.ai/case-study-modernization-three-setups.md (sources/case-study-modernization-three-setups.md L37 (under "In short"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 d454c4975cd1fbf90ebf353a0af6b18fde2926e2bf47abc19a0b54601f20e467
    Per-run cost table of Claude Code's own total_cost_usd: arm means $16.92 (A) and $10.04 (B′), B′ 41% below A.
    C-027Six checks were generated from the demo's own CALM architecture file and one was written by hand.onedroid.ai/llms-full.txt (sources/llms-full.txt L3406 (PAGE /case-study-finos-calm; under "In short"))
    onedroid.ai/partners-financial-services (sources/partners-financial-services.html L333-334 (under "03The proof: a FINOS CALM trading demo"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L33 (under "In short"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 37e5df715c2f8ee3670704caebfb06a0b7fb17883450b02ee31fa3284e49a1b0
    The `argus calm import` output lists exactly six generated checks, the by-id check file says it was written by hand, and the certificate's sealed set counts 7 scenarios.
    C-028The guardrail refused the restricted symbols in a trade search, but returned a restricted trade in full when the MCP server was asked for it by id.onedroid.ai/llms-full.txt (sources/llms-full.txt L3408 (PAGE /case-study-finos-calm; under "In short"))
    onedroid.ai/partners-financial-services (sources/partners-financial-services.html L349-351 (under "03The proof: a FINOS CALM trading demo"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L35 (under "In short"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 37e5df715c2f8ee3670704caebfb06a0b7fb17883450b02ee31fa3284e49a1b0
    Per-check run report of a final-mode run: the by-id step failed because it returned the full LSE:VOD trade (instrument, price, quantity, status) where 'is restricted' was asserted, while the VOD, GME and AMC search denials passed; the 19:34 control-plane record shows the same pattern.
    C-031Before the run, OneDroid Argus read the image digests actually running and found exactly the three that the certification declared. To show the other side, a final run was started that declared a digest that was not running. It was refused before any check ranonedroid.ai/llms-full.txt (sources/llms-full.txt L3510 (PAGE /case-study-finos-calm; under "How anyone can check it later"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L137 (under "How anyone can check it later"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 37e5df715c2f8ee3670704caebfb06a0b7fb17883450b02ee31fa3284e49a1b0
    The certificate's artifact measurement reads 'matched', with the three running digests equal to the three declared; the frozen control-plane record of the refused final run has 0 checks run and the quoted refusal for an undeclared digest.
    C-032Two build runs, outside the sealed set, served up to 160 concurrent users with a 95th percentile of 106 ms or less and no errors.onedroid.ai/llms-full.txt (sources/llms-full.txt L3526 (PAGE /case-study-finos-calm; under "What this test did not cover"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L153 (under "What this test did not cover"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 37e5df715c2f8ee3670704caebfb06a0b7fb17883450b02ee31fa3284e49a1b0
    Two build-mode run records under a set hash other than the sealed one: the load-ramp record's top step is 160 sessions at p95 106 ms with every step's delivered ratio 1 (no step above 106 ms), and the 10-user load run has error rate 0.
    C-296OneDroid Argus ran here against a development control planeonedroid.ai/llms-full.txt (sources/llms-full.txt L3536 (PAGE /case-study-finos-calm; under "What this test did not cover"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L163 (under "What this test did not cover"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 37e5df715c2f8ee3670704caebfb06a0b7fb17883450b02ee31fa3284e49a1b0
    The run-driving scripts address the argus-dev control plane and the certificate's private-chain anchors are on argus-dev-poa.
    C-297With the tester, the coding agent cost half as much ($1.01 against $1.97 a run), because it stopped testing its own work.onedroid.ai/llms-full.txt (sources/llms-full.txt L3584 (PAGE /case-study-claude-code-alone-vs-tester; under "Claude Code alone or with an independent tester: the same quality, and where the money went"))
    onedroid.ai/case-study-claude-code-alone-vs-tester.md (sources/case-study-claude-code-alone-vs-tester.md L21 (under "Claude Code alone or with an independent tester: the same quality, and where the money went"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 759cf275abe53c1dfbf3d72efb7e8049dc67ca1593c0644e9f4b32f5ca15aeb1
    Per-run cost table cross-checked against Claude Code's own totals; mean usd of B-1..B-3 = $1.01 and of A-1..A-3 = $1.97 (-49%), with B's test-command counts lower.
    C-299A fourth run of setup A was not judged and is left out of every figure here.onedroid.ai/llms-full.txt (sources/llms-full.txt L3626 (PAGE /case-study-claude-code-alone-vs-tester; under "The setup"))
    onedroid.ai/case-study-claude-code-alone-vs-tester.md (sources/case-study-claude-code-alone-vs-tester.md L63 (under "The setup"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 759cf275abe53c1dfbf3d72efb7e8049dc67ca1593c0644e9f4b32f5ca15aeb1
    Per-run table has an A-4 row with empty judge columns, and the page's arm-A figures (2.37 million tokens, 3 test runs) are the A-1..A-3 means, not the means with A-4.
    C-302List prices for Claude Opus 5.5: $4 input and $20 output per million tokens, $0.20 cache read, $5 cache write, $8 for a one-hour cache write.onedroid.ai/llms-full.txt (sources/llms-full.txt L3628 (PAGE /case-study-claude-code-alone-vs-tester; under "The setup"))
    onedroid.ai/case-study-claude-code-alone-vs-tester.md (sources/case-study-claude-code-alone-vs-tester.md L65 (under "The setup"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 759cf275abe53c1dfbf3d72efb7e8049dc67ca1593c0644e9f4b32f5ca15aeb1
    Per-run token-by-type table: at $4 input, $20 output, $0.20 cache read and $8 one-hour cache write every recorded run cost is reproduced exactly, and repricing the cache write at $5 reproduces the cross-check's lower Talos figures exactly.
    C-303With the tester, the coding agent ran its own tests twice in its three runs together; alone, it ran them 3 times in an average run.onedroid.ai/llms-full.txt (sources/llms-full.txt L3598 (PAGE /case-study-claude-code-alone-vs-tester; under "In short"))
    onedroid.ai/case-study-claude-code-alone-vs-tester.md (sources/case-study-claude-code-alone-vs-tester.md L35 (under "In short"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 759cf275abe53c1dfbf3d72efb7e8049dc67ca1593c0644e9f4b32f5ca15aeb1
    Per-run test_runs column (builder test-command calls); B-1..B-3 sum to 2 and A-1..A-3 average 3.
    C-304Every B run passed the tester in its first round.onedroid.ai/llms-full.txt (sources/llms-full.txt L3658 (PAGE /case-study-claude-code-alone-vs-tester; under "What the numbers say"))
    onedroid.ai/case-study-claude-code-alone-vs-tester.md (sources/case-study-claude-code-alone-vs-tester.md L95 (under "What the numbers say"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 759cf275abe53c1dfbf3d72efb7e8049dc67ca1593c0644e9f4b32f5ca15aeb1
    Per-run Argus columns: each B run has one completed Argus run with status passed and an Argus run id; the bundle's review reads 'first round' as the first completed run.
    C-307A tester ran 100 checks written from the library's user guide and returned only the names of the failures, for up to 4 rounds.onedroid.ai/llms-full.txt (sources/llms-full.txt L3746 (PAGE /case-study-modernization-three-setups; under "The setup"))
    onedroid.ai/case-study-modernization-three-setups.md (sources/case-study-modernization-three-setups.md L55 (under "The setup"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 d454c4975cd1fbf90ebf353a0af6b18fde2926e2bf47abc19a0b54601f20e467
    Harness code caps the tester at 4 rounds and writes only failed names as feedback; B's per-round names files hold check ids only, its round records show 4 rounds on the 100-check set, and the original itself fails some of those checks, as checks written from the guide rather than from its answers would.
    C-308the tester ran 106 equivalence checks against the original's recorded answers, half of a set of 215onedroid.ai/llms-full.txt (sources/llms-full.txt L3748 (PAGE /case-study-modernization-three-setups; under "The setup"))
    onedroid.ai/case-study-modernization-three-setups.md (sources/case-study-modernization-three-setups.md L57 (under "The setup"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 d454c4975cd1fbf90ebf353a0af6b18fde2926e2bf47abc19a0b54601f20e467
    Frozen split file, made before the first B′ run, lists 106 BUILD and 109 JUDGE names, disjoint, union 215; the per-run table records B′'s tester on the 106 BUILD set.
    C-309The other 109 equivalence checks, hidden from every setup.onedroid.ai/llms-full.txt (sources/llms-full.txt L3750 (PAGE /case-study-modernization-three-setups; under "The setup"))
    onedroid.ai/case-study-modernization-three-setups.md (sources/case-study-modernization-three-setups.md L59 (under "The setup"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 d454c4975cd1fbf90ebf353a0af6b18fde2926e2bf47abc19a0b54601f20e467
    None of the split file's 109 JUDGE names appears in any B or B′ feedback file (names lists and fix texts) and the per-run leak scan is 0 for every run; the record shows no observed access, not technical isolation.
    C-310Claude Code's own `total_cost_usd` per run. Our own count from the transcripts agrees within 0.11%.onedroid.ai/llms-full.txt (sources/llms-full.txt L3756 (PAGE /case-study-modernization-three-setups; under "The setup"))
    onedroid.ai/case-study-modernization-three-setups.md (sources/case-study-modernization-three-setups.md L65 (under "The setup"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 d454c4975cd1fbf90ebf353a0af6b18fde2926e2bf47abc19a0b54601f20e467
    Per-run cost table carries both the transcript-based count and Claude Code's total_cost_usd for every A, B and B′ run, largest gap 0.11%; the transcripts themselves are not in the bundle, so the own count is a recorded figure, not recomputable from it.
    C-311Cost per run | $16.92 | $15.19 | $10.04 (−41%)onedroid.ai/llms-full.txt (sources/llms-full.txt L3768 (PAGE /case-study-modernization-three-setups; under "The result"))
    onedroid.ai/case-study-modernization-three-setups.md (sources/case-study-modernization-three-setups.md L77 (under "The result"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 d454c4975cd1fbf90ebf353a0af6b18fde2926e2bf47abc19a0b54601f20e467
    Per-run cost table; per-arm means of Claude Code's total_cost_usd = $16.92 (A), $15.19 (B), $10.04 (B′), B′ 41% below A.
    C-313B′ stayed faithful and matched A exactly on the unambiguous hidden checks: 101 of 104.onedroid.ai/llms-full.txt (sources/llms-full.txt L3780 (PAGE /case-study-modernization-three-setups; under "What the numbers say"))
    onedroid.ai/case-study-modernization-three-setups.md (sources/case-study-modernization-three-setups.md L89 (under "What the numbers say"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 d454c4975cd1fbf90ebf353a0af6b18fde2926e2bf47abc19a0b54601f20e467
    Judge outputs restricted to the 109 JUDGE names minus the 5 start-sensitive lines listed in README before B′ ran give 101 of 104 for every A and every B′ run, with the same 3 failing names.
    C-315A block on compound shell commands didn't work in any B or B′ run: it failed open because a helper tool was missing.onedroid.ai/llms-full.txt (sources/llms-full.txt L3796 (PAGE /case-study-modernization-three-setups; under "What this experiment did not cover"))
    onedroid.ai/case-study-modernization-three-setups.md (sources/case-study-modernization-three-setups.md L105 (under "What this experiment did not cover"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 d454c4975cd1fbf90ebf353a0af6b18fde2926e2bf47abc19a0b54601f20e467
    Per-run hook columns measured from the session streams show 0 blocks in all six B and B′ runs and the guard's missing-helper (jq) fail-open note in four of them; the other two recorded no Bash hook event.
    C-316One interface rule could be read two ways. It affects 5 hidden checks and explains B′'s 2-check lead over A.onedroid.ai/llms-full.txt (sources/llms-full.txt L3794 (PAGE /case-study-modernization-three-setups; under "What this experiment did not cover"))
    onedroid.ai/case-study-modernization-three-setups.md (sources/case-study-modernization-three-setups.md L103 (under "What this experiment did not cover"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 d454c4975cd1fbf90ebf353a0af6b18fde2926e2bf47abc19a0b54601f20e467
    README lists the 5 start-sensitive CHKDER lines in the hidden half before B′ ran; judge outputs show A failing 2 of them and B′ none, with identical results on the other 104, which is the whole 2-check lead.
    C-333Claude Code ported a decades-old Fortran library to Goonedroid.ai/llms-full.txt (sources/llms-full.txt L3705 (PAGE /case-study-modernization-three-setups; under "Same modernization, three setups: what the tester checks decides whether it helps"))
    onedroid.ai/llms.txt (sources/llms.txt L61 (under "Docs"))
    onedroid.ai/case-study-modernization-three-setups.md (sources/case-study-modernization-three-setups.md L14 (under "Same modernization, three setups: what the tester checks decides whether it helps"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 d454c4975cd1fbf90ebf353a0af6b18fde2926e2bf47abc19a0b54601f20e467
    Every run has a recorded claude -p command and a judge output against its served Go port, and the DPMPAR diffs are edits to the ports' Go source; the port sources themselves are not in the bundle.
    C-619Two certified runs on one cluster the same day, and no capacity claim.onedroid.ai/llms-full.txt (sources/llms-full.txt L3395 (PAGE /case-study-finos-calm; under "Does the guardrail hold on every path? Seven sealed checks against a FINOS CALM demo"))
    onedroid.ai/case-study-finos-calm.md (sources/case-study-finos-calm.md L22 (under "Does the guardrail hold on every path? Seven sealed checks against a FINOS CALM demo"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 37e5df715c2f8ee3670704caebfb06a0b7fb17883450b02ee31fa3284e49a1b0
    Two final-mode run records dated 7 October 2026 (the certified 19:58 run and the 04:39 run, both 5 passed and 2 failed) against the one calm-demo instance; 'no capacity claim' is a disclaimer, consistent with the config's declared capacity gap.
    C-706An AI rewrite passed nearly all the checks written from the manual and still gave wrong numbers. Those only showed up against the original.onedroid.ai/llms-full.txt (sources/llms-full.txt L3837 (PAGE /case-study-three-ways-to-test-code; under "Three ways to test code you did not write"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L22 (under "Three ways to test code you did not write"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 d454c4975cd1fbf90ebf353a0af6b18fde2926e2bf47abc19a0b54601f20e467
    Per-run score table and per-run judge outputs: each rewrite gets the original's verdict on every behaviour check from the manual, while the equivalence checks against the original's recorded answers fail on wrong numbers the manual checks did not catch.
    C-715A rewrite of part of the program in another language got the same requests as the original.onedroid.ai/llms-full.txt (sources/llms-full.txt L3872 (PAGE /case-study-three-ways-to-test-code; under "3. Does an AI rewrite do what the original does?"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L57 (under "3. Does an AI rewrite do what the original does?"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 d454c4975cd1fbf90ebf353a0af6b18fde2926e2bf47abc19a0b54601f20e467
    Judge output files for the original and for each rewrite name the same equivalence scenario set and request count, and the results record states the reference and every candidate are called with identical requests.
    C-717The checks written from the manual alone passed nearly all of the rewrite.onedroid.ai/llms-full.txt (sources/llms-full.txt L3876 (PAGE /case-study-three-ways-to-test-code; under "3. Does an AI rewrite do what the original does?"))
    onedroid.ai/case-study-three-ways-to-test-code.md (sources/case-study-three-ways-to-test-code.md L61 (under "3. Does an AI rewrite do what the original does?"))
    Held by OneDroid, shown to us on 2026-11-05, shared on request.
    sha256 d454c4975cd1fbf90ebf353a0af6b18fde2926e2bf47abc19a0b54601f20e467
    Per-run manual-check score table and judge output: the rewrites fail only the checks the original itself fails plus one source-manifest pin, i.e. they pass nearly all of the manual set.

    Sources as fetched

    Every verdict above points into one of these files. To check one, fetch the URL with a browser user-agent and run sha256sum on the bytes; a different fingerprint means the page changed after 2026-11-07.

    SourceHTTPsha256 as fetched
    https://onedroid.ai/llms-full.txt2002b063c3ac9cf2b36575c490a05d064b2b280a96d5f0160af6285ca40966d2567
    https://onedroid.ai/llms.txt200b3355cab0e613148a95d9640e19ada8edbd2e9bb3e25012dc95ab87a36358088
    https://onedroid.ai/argus.md2000575679bceb60ccafedb77eb09d33ee1d130d333c9935fb96fad0eafab1cbd82
    https://onedroid.ai/argus-agents.md200ad0762f8785885258c083d573fcfcbcdb0922f33afcb21970d3bec55b1ced04c
    https://onedroid.ai/pricing.md2005ff12ee9325466144dab7b697ef5dc621b551ba40f1fc86c44f1e017014dbff5
    https://raw.githubusercontent.com/OneDro1d/argus-runner/main/README.md2001c7c100e35078b36c7a4fa26aa7c85c718a38c7749354fcf88c6f2298143da3a
    https://docs.onedroid.ai/argus-quickstart20092ec3bb9c0ab65930d77462e092914bee14425ef3805a5becdf6a601376e5121
    https://docs.onedroid.ai/argus-quickstart.md20001bc06254c0db57308743ea60288083a07e7e5b65bd547c4b8832304980bdf45
    https://docs.onedroid.ai/argus200c54d733672965cbe7eae8bdb13603bccfb84a187ebe428e550cac25bd53f6737
    https://docs.onedroid.ai/argus.md200e7dddb401cc4704d34db08fe46511337caaf32db52f2bb40aafad147ae24caaf
    https://api.github.com/repos/OneDro1d/argus-runner200849957978d1937cb2e3cea42a0c296b53417c0d44611de7371822e19d656212f
    https://onedroid.ai/financial-services.md200bf5f65ab6d90ab903b5b8d1f0a80cd76c1ad39567805d4af02ffdbc1d4c44736
    https://onedroid.ai/partners-financial-services200957d1a80d034232c697e44798b2286b0bcae73019c84c065943f9d07b774aba1
    https://onedroid.ai/case-study-otrs-znuny.md20095ff01719a47ad4d2d419b4127a99dcf62876e246f44499923309a9a6cc29cbc
    https://onedroid.ai/case-study-finos-calm.md200587377cd0d091a0f6830861614ec0a45bcd133ebcab72e414f85938fa4ad46ef
    https://onedroid.ai/case-study-claude-code-alone-vs-tester.md200931a53a9fbb839eaaaf9fcc50b4852bb1df1189fe2d35b82121df61f7b16ea49
    https://onedroid.ai/case-study-modernization-three-setups.md20070c5cb614e88878b93a6ea01728cfeb723ef3a45a0ba0dbc29bf82cbb65fcdd4
    https://onedroid.ai/case-study-three-ways-to-test-code.md200fe8001a53c8705c7dc443d9e35bfd99c412beff4ef5b28dd6b2312d1ad1f4cc3
    https://onedroid.ai/security.md20008b2a6382ad393864b405eab5cac62c30532471d151bdfb02291c0f240caef3f
    https://onedroid.ai/privacy.md2005fbcd6f0e1bf80d3290aba6b18b8f6dc2e7f3a4fe1b4a1c7b12dd73443e35b5f
    https://onedroid.ai/changelog.md2009020dbf8e2f1e79f6c8d977e00f176617b7c8a2ddfc2393efe4d48deb16b329f