[Deicyde]

Release-affecting (site/ops/brain-nightly.sh). Merge it just before release B is rebuilt, not between a build and its promotion. If it lands after B is built, the promoter refuses B. B already needs a rebuild at 905c6def or later anyway.

Change

The curated-input check (#83) exempts the fold's outputs with fnmatchcase(path, "catalog/data/*_links.jsonl"). fnmatch's * crosses /, so the exemption also covered any curated input below catalog/data, such as a future catalog/data/external/<x>_links.jsonl. The fold writes only catalog/data/<key>_links.jsonl, one per frontier source key in catalog/data/source_registry.json. The exemption now also requires no / after catalog/data/ (found in #83's re-review).

Nothing matches differently today: catalog/data/tauceti_links.jsonl is the only curated *_links input, and it stays exempt.

Tests (site/ops/test_brain_nightly.py)

• The behavioural test adds a curated catalog/data/external/x_links.jsonl and expects a change to it to be listed, next to the existing bot/state cases.
• The pin test checks the new condition.

I ran the changed snippet by hand: tauceti_links and fc_links are exempt, while catalog/data/external/x_links.jsonl and bot/state/cut_log.json are checked. bash -n passes under bash and /bin/bash 3.2. I didn't run the test suite locally because machine load was above the agreed limit; CI runs it.

[Deicyde]

Recommendation: merge just before release B is rebuilt. It is release-affecting. An adversarial review found no defect, and CI is green (25 nightly tests).

• Verified:
  • Old and new is_fold_output classify every curated input in all four inventories identically. Today's exemptions, including tauceti_links.jsonl, are unchanged.
  • The fold writes only direct children: REPO_KEY_RE admits no /.
  • The block parses and passes its heredoc through byte-identical under /bin/bash 3.2.
  • The behavioural test fails on main's code (the external x_links change goes unlisted) and passes here.
• Changed after review: the comment moved from the pattern line to is_fold_output, since the pattern alone still crosses /. The first version of that comment had an apostrophe, which bash -n showed breaks the script inside the $( … ) heredoc. The pushed version has none, and parses under bash and /bin/bash 3.2.