[Jammy2211]

Summary

The audit #267 asked for and never got. #268 fixed the PyAutoNerves half of the
Refactor Agent's witness map; this checks the rest, and Nerves turns out not to
have been a one-off.

test_witness had no row at all for PyAutoCTI or for five organs
(Brain, Heart, Hands, Memory, Mind) that all have real suites — so
behaviour_preservation reported them unwitnessed and the conductor advised
"strengthen tests first" for repos that are already tested. Two more carried
the same two-keys-per-repo defect #268 fixed for Nerves: @PyAutoCTI/@autocti
and @PyAutoReduce/@autoreduce normalised to different keys with nothing
joining them, and the Reduce row was keyed on the prefixed spelling — so the
bare package spelling, the one prompts actually use, missed the witness.

Repo	Test dir (verified)	Before
PyAutoCTI	test_autocti	missing entirely
PyAutoReduce	test_autoreduce	keyed pyautoreduce; @autoreduce missed
PyAutoBrain / Heart / Hands / Memory / Mind	tests	missing
PyAutoGut	none	correctly absent — now says so in a comment

Every row was verified by reading the repository's own tree
(PyAutoLabs/<repo>@main), not inferred from the naming convention — which
matters, because the organs break the test_<package> pattern the libraries
follow.

The keying rule is now stated where the map lives: canonical key is the bare
package name where the repo ships one, the repo name where it does not.

API Changes

None. Internal policy maps only; test_witness has exactly one consumer
(_refactor.behaviour_preservation) and this change only ever adds resolvable
keys — no repo that already resolved changes behaviour.

Test Plan

• python3 -m pytest tests/ -q → 468 passed
• python3 PyAutoMind/scripts/repos_sync.py --check → all 11 checks OK
• Live resolution check — all twelve tested repos now resolve under both
  spellings, unwitnessed_repos == ['pyautogut'] and nothing else
• Mutation-tested the guards rather than trusting them (below)

Three guards, and proof each one bites

All three derive from the body map, so they hold for an adopting fork and name
no instance fact — the tenant-firewall leg of repos_sync --check enforces
that, and it rejected two earlier drafts of these tests that hardcoded repo
names.

1. Coverage — every library/organ in the body map needs a witness row or
   a reasoned WITNESS_EXEMPT entry; an exemption naming a repo the body map
   has since dropped is itself a failure, so the allowlist can't rot into a
   blanket waiver.
2. Key sanity — a row's key must be what the repo it names normalises to.
   This is the invariant autoconf: PyAutoConf/test_autoconf violated while
   looking entirely plausible.
3. Both spellings — where a row's value shows the repo ships a package
   (<Repo>/test_<package> does, <Repo>/tests does not), both spellings must
   reach the same key.

Verified by mutation, not assertion:

Mutation	Caught by
drop the autocti witness row	guard 1
drop the pyautoreduce alias	guards 1, 2
restore the original mis-keyed pyautoreduce row	guards 1, 2, 3

That last row is the point: the pre-existing defect this PR fixes would now
fail CI three ways.

One gap found and deliberately left open

The bare spellings of the organ repos still split: _target_sets registers
both pyautobrain and autobrain as known targets, but only the prefixed form
is canonical, so @autobrain, @autoheart, @automemory and @automind
resolve to keys nothing is filed under. I fixed only @autohands, because
extra_organism_targets explicitly declares it and it therefore reaches real
code today; the others are latent and fixing them means deciding whether organs
should key bare or prefixed across target_signals too — a policy question
wider than this prompt. Filed as a note on #269 rather than widened into this
PR.

Reviewer caveats

• Heart was not consulted — pyauto-heart is unreachable in this
  web-github session, same as fix: retire PyAutoConf rename leftovers in Brain functional surfaces #268. The suite stood in per the WORKFLOW.md
  fallback and is green, but no verdict was obtained.
• Review-faculty claim disposition — the surface lifted "Every row was
  verified against the repository itself": basis-cited, by
  git ls-tree HEAD over blobless clones of all ten repos, which is also what
  produced the table above. The other lifted claim is fix: retire PyAutoConf rename leftovers in Brain functional surfaces #268's and is
  dispositioned on #267.

Ordering

Stacked on #268 — same file, same two maps. Based against
claude/pyautoconf-rename-leftovers-38ia48 so the diff shows only this task's
change; it will retarget to main automatically when #268 merges. Merge #268
first.

Effective autonomy safe (header safe, maintenance cap safe); plan
written to #269 at the start, unmodified since. Gate: tests PASS (468) · smoke
n/a (organism repo) · review CLEAN with the disposition above · Heart not
evaluated.

Closes #269

🤖 Generated with Claude Code

https://claude.ai/code/session_01MPhEXMoPYa7H1sNHWR2GNW

---

Generated by Claude Code