New WSL distro recipe follow-up: rehearsal, signed sums, schema and storage pre-checks, ownership probe; three corrected facts - #582

… merged (#582)

- cross:wsl-distro no longer says stage 1 waits for the follow-up: PR #582 merged as b8dd81d, and the recipe now
  starts with a rehearsal on a throwaway name and pre-checks in the workstation distribution. The install command,
  the new-host steps, the lane gate and the notes say so; nothing in the recipe has run on a host.
- The winner's pin_source follows the pinned hash to its new line in the recipe (250, was 99): the follow-up
  inserted sections above it, and the build checks a citation's bounds, not its content.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

… merged (#582)

- cross:wsl-distro no longer says stage 1 waits for the follow-up: PR #582 merged as b8dd81d, and the recipe now
  starts with a rehearsal on a throwaway name and pre-checks in the workstation distribution. The install command,
  the new-host steps, the lane gate and the notes say so; nothing in the recipe has run on a host.
- The winner's pin_source follows the pinned hash to its new line in the recipe (250, was 99): the follow-up
  inserted sections above it, and the build checks a citation's bounds, not its content.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

… merged (#582)

- cross:wsl-distro no longer says stage 1 waits for the follow-up: PR #582 merged as b8dd81d, and the recipe now
  starts with a rehearsal on a throwaway name and pre-checks in the workstation distribution. The install command,
  the new-host steps, the lane gate and the notes say so; nothing in the recipe has run on a host.
- The winner's pin_source follows the pinned hash to its new line in the recipe (250, was 99): the follow-up
  inserted sections above it, and the build checks a citation's bounds, not its content.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

…e); keys row, landed sources, distribution row (#583)

* Architecture edition: the keys lane's row after the canary proof merged; #578's sources as plain citations

The first update of the dated edition under its own overturn conditions 4 and 5.

- cross:credential-practice, in the keys lane owner's wording: the canary proof tool merged with #579 (e58850f)
  and its synthetic acceptance on the workstation is recorded (143 unit tests, 138 of 138 mutants, 531 suite tests
  with one failure by design); the acceptance names what the two receipts record, class local_integration_check;
  the end-to-end proof on the new distribution has not run on any host and stays a new-host step; closure c3 and c4
  reworded.
- The seven citations of PR #578, which merged as 65a7b03, become plain source paths; the record lists five merged
  pull requests and two open ones (#575, #535).

Co-Authored-By: Claude Fable 5.1 <[email protected]>

* Architecture edition: the distribution row after the recipe follow-up merged (#582)

- cross:wsl-distro no longer says stage 1 waits for the follow-up: PR #582 merged as b8dd81d, and the recipe now
  starts with a rehearsal on a throwaway name and pre-checks in the workstation distribution. The install command,
  the new-host steps, the lane gate and the notes say so; nothing in the recipe has run on a host.
- The winner's pin_source follows the pinned hash to its new line in the recipe (250, was 99): the follow-up
  inserted sections above it, and the build checks a citation's bounds, not its content.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

* Architecture edition: selections follow the repositories' own quality, not the source host's legacy

The user's rule of 2026-10-01: evaluate each repository on its own quality and do not shape the new distribution by
what the source host has installed, pinned or failed to integrate.

- durable-memory: ai-memory is the reference arm, not a default (the 2026-09-27 decision). agentmemory joins with
  the one matched harness result on record (recall_all@5 0.821 against 0.570 on LongMemEval-S, Mac, descriptive),
  MemPalace as an arm, Hindsight as arm K1 judged fresh; the gate no longer treats this host's integration holds as
  evidence about the repository; the first new-host step installs every arm fresh.
- token-efficiency, in the Gate A owner's words: comparison_required; the 14-component profile is the source
  host's selection and one arm, the lean base an arm of equal standing, and the E2E decides which components stay.
- Trading rows, in the trading lane owner's words: the verdict selections that had been listed as alternatives for
  lacking a stack pin are winners of their layers (DVC, pandera, agent-retrieval-bench, Inspect AI, MLflow), each
  with its upstream install command and the class its recorded run supports.
- cross:runtime-workers: a candidate without a repository pin is a candidate, not an exclusion.
- Every comparison row carries the reason that its merit winner is undetermined and that the comparison selects.

Verdicts: 20 selection_of_record_open, 15 comparison_required, 2 new_host_required. 130 winner entries.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

* Architecture edition: the memory comparison is against ai-memory with its reranker off, not production

A cross-family audit corrected the wording of the one matched memory result. The 0.821 against 0.570 recall on
LongMemEval-S compares agentmemory's arm D2 with ai-memory's arm C3: the production embedder and query prefix with
the reranker off. ai-memory's production arm runs the LLM reranker (C4, amendment A14) and has not run, nor has
agentmemory through its shipped hooks (D2h); the arms' captures and embedders differ and the ai-memory build was a
2.5 pre-release. The durable-memory row and the record now say configuration-level evidence, not a production
head-to-head, and cite the convergence record itself (paired difference +0.251, 95% CI +0.203 to +0.299).
Overstating a challenger is the same bias the merit rule excludes.

Co-Authored-By: Claude Fable 5.1 <[email protected]>

* Re-register this branch's changed files in manifests/evidence.json (hot-file protocol)

Co-Authored-By: Claude Fable 5.1 <[email protected]>

---------

Co-authored-by: Scout <scout@local>
Co-authored-by: Claude Fable 5.1 <[email protected]>