mellions-bug-auditlisted

<!-- Mellions Engineer | LetA Tech Ltd. | leta@letatech.ca --> # Auditing for defects Establish, from the executable implementation, whether a defect is real — what causes it, what it reaches, what evidence supports each claim. The output is a finding another engineer can act on without repeating the investigation: what must become true, never the diff and never an issue, which a confirmed finding becomes by `mellions-issue-creation`. ## What the ground truth is `mellions-deep-research` ranks the sources. For an audit that means behaviour is established from the implementation at a named commit, the tests that exercise it, execution and runtime observation — and where a document and the code disagree, the code is correct and the document is stale, said so in the finding, because a stale document nobody corrects produces the same wrong conclusion next quarter. **Where the artifact under audit is itself a process** — a doctrine, a canon, a policy, a skill — its ground truth is what the fleet actually does, and that is counted, not read: a document asserting a control is a claim that the control is operated. One estate's canon documented a ten-label lifecycle no registry defined and a gate whose permitting label had been applied once across 860 open issues — one query each, settling what re-reading the prose could not. ## Tier every claim Established, inferred and unknown, in the words a finding uses: **confirmed from source** (code at a named commit, `file:line` with the