Numerroo Assay publication control

Status: active. Adopted 2026-07-02. This document governs whether an Assay research candidate may move from frozen evidence to public release. It is an internal control, not legal advice.

Authority

No Assay output publishes because a model, test, draft, or reviewer says it is persuasive. Publication requires evidence integrity, claim discipline, owner approval, and any review required by the target class.

The owner is the final publication gate. An LLM may draft or challenge a piece; it cannot approve or publish it.

Release states

DRAFT
  -> EVIDENCE_FROZEN
  -> TECHNICAL_REVIEWED
  -> RIGHT_OF_REPLY_COMPLETED_OR_NOT_REQUIRED
  -> LEGAL_REVIEWED_OR_OWNER_NOT_REQUIRED
  -> OWNER_APPROVED
  -> PUBLISHED
  -> CORRECTED or RETRACTED

A state may advance only when its receipt records who decided, when, and on what evidence. NOT_REQUIRED is a documented decision, not a skipped field.

Target classes

No piece may imply that internal policy review is legal clearance.

Evidence contract

Every publication candidate must carry a stable receipt containing:

1. a unique receipt ID and assay version;
2. target scope and the exact claim under assay;
3. immutable input files or a durable reference to their retained location;
4. SHA-256 for every input, generated fill set, and verdict payload;
5. a manifest hash and an engine commit SHA;
6. complete parameters, cost assumptions, data window, and venue/path assumptions;
7. exact reproduction commands, including every non-default flag relied upon;
8. reconciled headline figures and any derived calculations;
9. known limitations, favourable assumptions, and unresolved uncertainty;
10. the review state and decision log.

Future evidence manifests use relative filenames and include the engine-commit receipt. Assay 001 predates that rule: its original manifest contains absolute Mainframe paths and omits keel_commit.txt. Its checked-in evidence receipt preserves the basename/hash mapping and records that limitation rather than rewriting the frozen snapshot.

Claim discipline

• State what was measured, under which parameters, over which window.
• Distinguish a mechanism finding from a product, person, or market-wide finding.
• Distinguish completed-cycle profit, marked inventory, and whole-account return.
• Do not use motive language or labels such as scam, fraud, deceptive, worthless, or safe.
• Do not convert a single path into a universal forecast.
• Do not hide strategy-favourable findings, failed counterexamples, or limitations.
• Do not use a disclaimer to contradict the overall impression of the piece.
• Do not publish a number that cannot be traced to the frozen receipt.

Review checklist

A candidate cannot advance beyond TECHNICAL_REVIEWED until:

• every manifest entry verifies;
• the engine commit exists and is recorded in full;
• every table value reconciles to the receipt;
• calculations and rounding are independently checked;
• no unresolved placeholder remains;
• the verdict language matches the sanctioned template;
• source-path, fee, spread, slippage, and depth assumptions are explicit;
• the result is presented with its strongest material limitation;
• the reproduction block is complete.

Owner approval additionally confirms headline, tone, target classification, publication channel, right-of-reply decision, and whether professional review is required.

Right of reply

For a named target, record:

• the exact findings supplied;
• the questions asked;
• delivery method and date;
• response deadline;
• response received or no response;
• how the response changed the piece.

Publish the response verbatim where practical or fairly summarise it. Preserve the full response privately. A response does not control the verdict, but material factual corrections must be tested before release.

Corrections and retractions

Errors are corrected promptly and visibly. The public page must show the correction date, the original affected statement, the corrected statement, and whether the verdict changed. The prior version and evidence remain retained.

Retract when the evidence package cannot support the central claim, the wrong target was tested, or a material integrity failure makes correction insufficient. Retraction is not deletion: the notice and reason remain public.

Register

Assay 001 decision log

• 2026-07-03 — Owner editorial review completed on draft v0.4. Headline, tone, target classification (category/open-source mechanism), and publication channel accepted.
• 2026-07-03 — Professional legal review recorded as not required for this candidate: a category-level mechanism assay with no vendor, product, interface, or person named. This is an internal policy decision and is not legal clearance.
• 2026-07-03 — Register advanced TECHNICAL_REVIEWED -> OWNER_APPROVED. Publication remained a separate later gate.
• 2026-07-03 — Release candidate deployed at https://assay.numerroo.com.au; DNS resolved, HTTPS returned HTTP/2 200, and CSP, HSTS, nosniff, frame-deny and referrer-policy headers were present.
• 2026-07-03 — Live content verification matched the approved verdict figure, evidence-engine commit and original manifest SHA-256. The pre-publication indexing block was confirmed before closeout.
• 2026-07-03 — Owner authorised go-live closeout. Register advanced OWNER_APPROVED -> PUBLISHED; indexing enabled.

Corrections register