ci(timings): give the test-timing parser self-test teeth in lint.yml, not continue-on-error in ci.yml

`scripts/report-test-timings.mjs` ships a real self-test — 61 cases across 4
batteries, a naive prefix-only reference parser plus three controls that assert
it gets the answer wrong exactly where the real parser gets it right. ci.yml
named it, so `check:self-test-wired` was green. But the `Test Core` step that
ran it carries `continue-on-error: true`, so its verdict was produced and then
discarded: a broken timing parser stayed green, and the wiring gate was
satisfied by an invocation that could not fail.

That `continue-on-error` cannot simply be removed. #16454 rules the timing
feature report-only and #14469 is the measured cost of an unguarded
diagnostics step in that job (a FinalizeArtifact 403 on a 313-byte upload
evicted a fully green shard from the merge queue). Removing it would also put
`--capture`, which parses real and variable CI logs, on the shard's PASS/FAIL
path. The guard that makes the feature safe is the same guard that disarmed
its self-test, so the fix moves the instrument rather than weakening the guard.

- lint.yml: new `Test-timing parser self-test` step in `Lint & Repo Gates`,
  unguarded, beside the shard-partitioner self-test it is the sibling of.
- ci.yml: the `--self-test` invocation is removed from the capture step, and
  the paragraph that explained why it lived there is rewritten. `--capture`
  and `--merge` keep `continue-on-error`; that property is untouched.
- report-test-timings.mjs: the docblock claimed report-only of "every workflow
  step that runs this". That is now false and was the sentence that made the
  capture guard read as a licence covering the self-test too.

`check:self-test-wired` stays green with the invocation moved: lint.yml is a
workflow, and the gate credits any workflow that names the script.

Claude-Session: https://claude.ai/code/session_012GKcPZbMoGq7WPzKLfRBTU
Co-authored-by: Claude <noreply@anthropic.com>

Author: claude

.github/workflows/ci.yml

@@ -928,12 +928,17 @@ jobs:
       # here is fine; moving them below that pair is not — and ⛔ that pair does
       # not acquire an `if:` or a `continue-on-error:` to "match" these.
       #
-      # The `--self-test` runs here because this is where CI names the script,
-      # which is what check:self-test-wired requires. ⚠ Under
-      # `continue-on-error` its verdict is VISIBLE but not enforcing: inside
-      # this workflow, no step this card may add is permitted to fail anything.
-      # Giving that self-test teeth means a `check:` step in lint.yml, which is
-      # deliberately NOT taken here.
+      # ⛔ The `--self-test` is deliberately NOT invoked here (#17097). It ran
+      # in this step until then, where `continue-on-error` discarded its verdict:
+      # the self-test reached an answer nothing could act on, and
+      # `check:self-test-wired` was satisfied by an invocation that could not
+      # fail. A self-test with no teeth is worse than no self-test, because its
+      # presence is what stops anyone from adding a real one. It now runs in
+      # lint.yml's `Lint & Repo Gates` job, as `Test-timing parser self-test`,
+      # with no `continue-on-error` — so a broken timing parser reddens there
+      # while this workflow keeps the report-only property #16454 ruled and
+      # #14469 measured the cost of losing. ⛔ Do not re-add it here; giving it
+      # teeth in `ci.yml` is exactly the trade-off those two cards refuse.
       - name: Capture this shard's test timings
         if: always()
         continue-on-error: true
@@ -942,7 +947,6 @@ jobs:
             echo "No test log — the test step did not get far enough to produce one."
             exit 0
           fi
-          node scripts/report-test-timings.mjs --self-test
           node scripts/report-test-timings.mjs --capture \
             --log "$RUNNER_TEMP/test-core.log" \
             --summaries .turbo/runs \

.github/workflows/lint.yml

@@ -3943,6 +3943,31 @@ jobs:
       - name: Shard partitioner self-test
         run: node scripts/partition-test-shards.mjs --self-test
 
+      # A self-test that RUNS but cannot fail anything (#17097) — the sibling of
+      # the defect above, one turn further out. `scripts/report-test-timings.mjs`
+      # ships a real self-test: 61 cases across 4 batteries, a naive prefix-only
+      # reference parser plus three controls that assert it gets the answer WRONG
+      # exactly where the real parser gets it right. ci.yml names it, so
+      # `check:self-test-wired` was green. But every `Test Core` step that runs
+      # that script carries `continue-on-error: true`, and MUST: #16454 rules the
+      # timing feature report-only, and #14469 is the measured cost of an
+      # unguarded diagnostics step there (a FinalizeArtifact 403 on a 313-byte
+      # upload evicted a fully green shard from the merge queue). So the guard
+      # that makes the feature safe was the same guard that disarmed its
+      # self-test — the verdict was produced and then discarded, and the wiring
+      # gate was satisfied by an invocation that could not fail.
+      #
+      # ⇒ the self-test runs HERE, where it has teeth, and ⛔ is no longer
+      # invoked from ci.yml: ONE enforcing run, not one enforcing and one
+      # decorative. ci.yml's own `continue-on-error` on the capture and merge
+      # steps is untouched by this — the report-only property belongs to that
+      # workflow, not to this script.
+      # Invoked as `node` rather than through a `pnpm check:*` alias for the same
+      # reason as the steps above: see the GATE INVOCATION IDIOM note at the top
+      # of this file. In-process fixtures, no IO, milliseconds.
+      - name: Test-timing parser self-test
+        run: node scripts/report-test-timings.mjs --self-test
+
       # Hand-written `.d.mts` mirrors (#10549). `scripts/js-comment-mask.mjs`
       # and `scripts/check-regen-pending.mjs` are untyped `.mjs` that each ship
       # a hand-written declaration beside them, and both files say "keep this in

scripts/report-test-timings.mjs

@@ -18,11 +18,21 @@
  * summaries (`.turbo/runs/*.json`, uploaded per shard) carry per-PACKAGE
  * execution windows and nothing finer.
  *
- * REPORT ONLY. Nothing here may redden `Test Core` -- that is the required
- * branch-protection context, and a diagnostics step that can fail it buys a
- * measurement at the price of the merge queue. Every workflow step that runs
- * this carries `if: always()` and `continue-on-error: true`, and the tool
- * itself degrades to a named refusal rather than throwing.
+ * REPORT ONLY IN `ci.yml`. Nothing in that workflow may redden `Test Core` --
+ * that is the required branch-protection context, and a diagnostics step that
+ * can fail it buys a measurement at the price of the merge queue. Every
+ * `ci.yml` step that runs this carries `if: always()` and
+ * `continue-on-error: true`, and the tool itself degrades to a named refusal
+ * rather than throwing.
+ *
+ * -- and that is a property of `ci.yml`, NOT of this script (#17097). This
+ * docblock used to claim it of "every workflow step that runs this", which
+ * made the `--capture`/`--merge` guard read as a licence covering the
+ * self-test too. `--self-test` runs in lint.yml's `Lint & Repo Gates` job with
+ * NO `continue-on-error`: a red verdict there is the whole point, and it is
+ * the only instrument watching the parsing rules below -- a clean tree cannot
+ * tell a working parser from a weakened one, which is what the naive-parser
+ * controls in the battery exist to say out loud.
  *
  * ## THE MEASUREMENT THAT CHOSE THE ROUTE (route (a): parse the stream)
  *