CLAUDE.md says 91 test targets; bazel query says 98, and did before the last branch #81

New issue

Closed

opened 2026-09-09 17:52:19 +00:00 by bcox · 1 comment

bcox commented

2026-09-09 17:52:19 +00:00

Owner

CLAUDE.md's Build and test section says:

bazel test //...                        # 91 test targets

bazel query 'tests(//...)' returns 98, and returned 98 at 1f09979 as well -- so the figure was already stale before the inter-layer-routing branch, which added test cases but no new targets.

Small, but this is the file agents read first to calibrate what a green run looks like, and a wrong count is the kind of thing that gets "corrected" in the wrong direction: during the inter-layer-routing work one agent had to be told explicitly not to adjust its expectations down to 91 when its own run reported 98.

It was left alone at merge time on purpose -- an unrelated documentation edit does not belong in a feature diff -- which is why it is here instead.

Context: observed during inter-layer routing, merged at 8422929.

`CLAUDE.md`'s Build and test section says: ```sh bazel test //... # 91 test targets ``` `bazel query 'tests(//...)'` returns **98**, and returned 98 at `1f09979` as well -- so the figure was already stale before the inter-layer-routing branch, which added test cases but no new targets. Small, but this is the file agents read first to calibrate what a green run looks like, and a wrong count is the kind of thing that gets "corrected" in the wrong direction: during the inter-layer-routing work one agent had to be told explicitly not to adjust its expectations down to 91 when its own run reported 98. It was left alone at merge time on purpose -- an unrelated documentation edit does not belong in a feature diff -- which is why it is here instead. Context: observed during inter-layer routing, merged at 8422929.

bcox referenced this issue

2026-09-09 23:56:08 +00:00

//patches:audition --patch is not repeatable, so an A/B needs two invocations #88

bcox referenced this issue

2026-09-09 23:56:18 +00:00

//patches:audition reports success after writing nothing #69

bcox referenced this issue

2026-09-09 23:56:25 +00:00

patches/audition.cc: stale bank count in the allocation comment #63

bcox commented

2026-09-09 23:56:35 +00:00

Author

Owner

Fixed on worktree-audition-flags, together with #88, #69 and #63.

The count is removed rather than corrected to 98. It is 99 as of this branch, which is the point: the number only rises, so any figure written here is a future version of this issue.

bazel test //...                        # the whole suite; all of it must pass

And the reasoning is now in the file, so nobody re-adds one:

The number of targets that runs is deliberately not recorded here. It only rises, and a figure in this file is something a run gets corrected against — an agent whose own run reported more than the number written here has already been talked out of trusting it. bazel query 'tests(//...)' | wc -l answers it for the tree in front of you.

That sentence is this issue's own observation — the agent during inter-layer routing that had to be told not to adjust down to 91 — kept where the next reader will hit it.

A second instance of the shape, in the same file, fixed with it: the --test_filter paragraph said "19 of this bank's 30 patch names are hyphenated." The bank is 40 (see #63), so that count was stale too, and it was load-bearing for the warning it supports. Now "most of this bank's patch names are hyphenated" — which keeps the force of the point and cannot rot.

This is the fourth instance of the class #63 tracks: CLAUDE.md's "29 patch names" (fixed on modal-cymbals), the EveryBlobsMacroKnobsAreItsOwn blob counts (same branch), audition.cc's allocation comment (#63) and this one.

Fixed on `worktree-audition-flags`, together with #88, #69 and #63. The count is removed rather than corrected to 98. It is 99 as of this branch, which is the point: the number only rises, so any figure written here is a future version of this issue. ```sh bazel test //... # the whole suite; all of it must pass ``` And the reasoning is now in the file, so nobody re-adds one: > The number of targets that runs is deliberately not recorded here. It only rises, and a figure in this file is something a run gets *corrected* against — an agent whose own run reported more than the number written here has already been talked out of trusting it. `bazel query 'tests(//...)' | wc -l` answers it for the tree in front of you. That sentence is this issue's own observation — the agent during inter-layer routing that had to be told not to adjust down to 91 — kept where the next reader will hit it. **A second instance of the shape, in the same file, fixed with it:** the `--test_filter` paragraph said "19 of this bank's 30 patch names are hyphenated." The bank is 40 (see #63), so that count was stale too, and it was load-bearing for the warning it supports. Now "most of this bank's patch names are hyphenated" — which keeps the force of the point and cannot rot. This is the fourth instance of the class #63 tracks: `CLAUDE.md`'s "29 patch names" (fixed on `modal-cymbals`), the `EveryBlobsMacroKnobsAreItsOwn` blob counts (same branch), `audition.cc`'s allocation comment (#63) and this one.

bcox referenced this issue from a commit

2026-09-10 00:09:21 +00:00

Make --patch repeatable; create the output directory it names

bcox closed this issue

2026-09-10 00:09:21 +00:00