docs(comments): give the a57 findings doc a findable path

[arkash20]

Three source comments — type_ii_materializer.py:9, organization_settings.py:76, recall_service.py:58 — cite benchmark/a57-recall-experiments-findings.md as if it were in this repo. It is not, and never has been on any ref.

It exists, at that path relative to the multi-repo workspace root, one level above this checkout. It is a local analysis artifact and is not published, so the citation resolves only if your working directory happens to be the workspace root — and is a dead pointer from inside a clone.

Why a comment-only change is worth making

Those comments name a convention and then point at nothing:

Change only with a fresh control pair on the 67-q regression sample (see benchmark/a57-recall-experiments-findings.md).

A reviewer on #1706 searched every ref for that file before asking where it was. The doc does carry what the comments promise — §"A64 control run — true-premise cost (2026-08-27, 67-q regression sample, d=0 fast)" has the method line and the per-category table, including the −14.3pp knowledge-update cost that is the reason the convention exists.

What each site now says

That the path is relative to the workspace root, and that a checkout alone will not find it.

recall_service.py additionally points at the regression-bench skill, because the second thing that reviewer was blocked on was believing the control pair needed hosted Caura credentials and a live instance. It does not: it runs against the local docker stack with dev-admin-key and takes about eight minutes at d=0. That misconception was about to gate a security fix behind infrastructure nobody needed.

(Also corrects the line number: the recall citation is at :58, not :105.)

Not done here

Publishing the doc itself. The honest options are to vendor the control-pair procedure into BENCHMARKS.md, or to move the file in after a scrub — parts of it reference internal benchmark scores, so it cannot be dropped into a public repo as-is. Worth its own decision rather than smuggling into a comment fix.

Checks

Comments only; no behaviour change. ruff check + format --check clean. -k "recall or premise or materializ or settings": 7 failed / 301 passed on this branch and the same 7 on a pristine origin/main worktree — identical sets, zero regressions. All 7 are FTS-dependent and share one cause: the test DB is built from the ORM models, so migration 001's search_vector trigger never exists and the column is NULL on every row. Pre-existing, filed separately, not touched here.

🤖 Generated with Claude Code

[github-actions]

Claude Code Review — skipped: PR author 'arkash20' is not a public member of the 'caura-ai' org

[arkash20]

@Eldad-Caura please approve

[github-actions]

Claude Code Review — skipped: PR author 'arkash20' is not a public member of the 'caura-ai' org

[github-actions]

Claude Code Review — skipped: PR author 'arkash20' is not a public member of the 'caura-ai' org