Skip to content

docs(audits): commit the 2026-08-04 audit report per ADR-0006 - #114

Closed
dlovell wants to merge 1 commit into
mainfrom
docs/audit-2026-08-04
Closed

docs(audits): commit the 2026-08-04 audit report per ADR-0006#114
dlovell wants to merge 1 commit into
mainfrom
docs/audit-2026-08-04

Conversation

@dlovell

@dlovell dlovell commented Aug 4, 2026

Copy link
Copy Markdown
Contributor

ADR-0006 places one mechanical obligation on the caller: commit each audit report to docs/audits/<date>-<scope>.md, because the auditor's "rerun and compare" promise is — in the ADR's own words — "vacuous while reports die in the invoking conversation."

I ran the audit of 88d16b4, acted on all three shapes (#109, #110, #112), and left the report in the conversation. That is precisely the failure the ADR exists to prevent. Without this, the M1–M9 measurement series, the seven mutation results, and the checked-and-dropped list all get re-derived from scratch next run.

Conformance after this PR:

ADR-0006 requirement Status
Auditor reads the newest prior report before measuring ✅ read 2026-08-02-adr-0005-acceptance.md, re-ran M1–M7
Agent writes no files
A disposal draft per shape
Caller commits the report to docs/audits/ this PR
Each shape filed, landed, or recorded #109, #110, #112

The report is verbatim apart from two caller additions:

  1. A disposition table, so the next reader sees which shapes closed and where.
  2. A correction block on Shape 3. Its "satisfied at introduction zero times in six" claim was falsified while reviewing docs: record two audit shapes as accepted, with reasons #112 — a selection effect, with feat: Nix devcontainer base — CI-published, multi-arch, default for non-seed overlays #41 and feat: track .claude/agents in git; per-subdir state symlinks (ADR-0003) #74 appearing in the report's own violation list while those same commits shipped at-introduction guards. Committing a known-false claim into the durable baseline would be worse than not committing it, so the correction ships with it, including the generalisable lesson: an "N for N" history claim needs the inverse query — are there contradicting instances outside the sampled set? — not just verification that the sampled ones are individually correct.

Docs only. pre-commit run --all-files clean.

🤖 Generated with Claude Code

ADR-0006 puts one mechanical obligation on the caller: commit each audit
report to docs/audits/<date>-<scope>.md, because the auditor's "rerun and
compare" promise is "vacuous while reports die in the invoking conversation."

The audit of 88d16b4 was run, all three of its shapes were dispositioned
(#109, #110, #112) — and the report itself was left in the conversation.
That is the exact failure the ADR exists to prevent, so the M1-M9 series,
the seven mutation results and the checked-and-dropped list would have been
re-derived from scratch next run.

The report is verbatim apart from two caller additions:
- a disposition table at the top, so the next reader sees which shapes closed
  and where;
- a correction block on Shape 3. Its "satisfied at introduction zero times in
  six" claim was falsified in review of #112: it was a selection effect, and
  #41 and #74 appear in its own violation list while those same commits
  shipped at-introduction guards. The correction records the lesson for
  future audits — an "N for N" history claim needs the inverse query (are
  there contradicting instances OUTSIDE the sampled set?), not just
  verification that the sampled ones are individually right.

Leaving the false claim uncorrected in the durable record would have been
worse than not committing it at all.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@dlovell

dlovell commented Aug 4, 2026

Copy link
Copy Markdown
Contributor Author

Closing — holding the audit report out of git for now at the author's request. The file stays as a working draft; ADR-0006's commit step will happen separately once reviewed.

@dlovell dlovell closed this Aug 4, 2026
@dlovell
dlovell deleted the docs/audit-2026-08-04 branch August 4, 2026 17:21
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant