A red check's summary names each shard's first failure, never the (unnamed) fallout behind it #8750

A red check's summary now gives each red shard's first named failure, with shards in shard order rather than the order they finished. The change is all in parseBunTestFailure (src/checks/bun-test.ts):

• Shard attribution: each (fail) line goes to a shard. It uses the [shard N] echo when the line has one, otherwise the ── shard N/M header above it, whether live or held. Output with no shards counts as a single shard.
• Fallout: a shard's (unnamed) entries, including those nameUnnamed has renamed, count only when that shard names no other failure.
• Order: the first name from each red shard comes first, in shard order, followed by each shard's other names in the order they were printed. The list is still capped at MAX_FAILING_NAMES. The count stays true, so leftovers show up as "and N more".

src/checks/summary.ts and src/checks/service.ts did not need changes, since both read this list. CheckJudgement.cause still tells environment faults apart as before: the full-disk mark comes from fullShard(lines) on the raw log, not from the names. rerunnable still refuses a re-run whenever the named failures don't add up to the failure count, so fallout can't make a red shard re-runnable.

Tests (test/checks.test.ts, in the parser suite):

• A shard whose named failure is followed by (unnamed) fallout names only the named failure.
• A shard with nothing but fallout is still named by it.
• Three shards logged in the order 3, 1, 2, with the tally echo after them, are listed in the order 1, 3.

One existing test changed: test/shared-red.test.ts (#4582, "summary names the file and the hook") expected a hook timeout followed by two (unnamed) entries. Under this issue's rule that shard has a named failure, so it now expects only a.test.ts > beforeEach/afterEach (timeout) and a summary ending "and 1 more". That test recorded the old behaviour, which this issue replaces. A side effect: an afterAll that throws in the same shard, which shows up as (unnamed): error: …, is no longer named in the summary, only counted. Its line is still in the log. Renamed hook timeouts count as named failures, not fallout, because the issue asks only about (unnamed).

Not done: no change to how scripts/test-shards.ts writes the log. Its final tally was already in shard order; the summary was taking its names from the per-shard sections, which are in finish order.

Checks run: bun scripts/gate.ts, bun x tsc --noEmit and the 39 test files that pass (fail), (unnamed) or checkDetail through the parser all pass (497 tests). scripts/fast.ts picked 268 files, about 24 minutes, so I ran those 39 instead and left the full run to the forge's check.

To try it: bun test test/checks.test.ts test/shared-red.test.ts test/disk-full-shard.test.ts

🤖 Generated with Claude Code

Closes #8730

A red check's summary now gives each red shard's first named failure, with shards in shard order rather than the order they finished. The change is all in parseBunTestFailure (src/checks/bun-test.ts):

• Shard attribution: each (fail) line goes to a shard. It uses the [shard N] echo when the line has one, otherwise the ── shard N/M header above it, whether live or held. Output with no shards counts as a single shard.
• Fallout: a shard's (unnamed) entries, including those nameUnnamed has renamed, count only when that shard names no other failure.
• Order: the first name from each red shard comes first, in shard order, followed by each shard's other names in the order they were printed. The list is still capped at MAX_FAILING_NAMES. The count stays true, so leftovers show up as "and N more".

src/checks/summary.ts and src/checks/service.ts did not need changes, since both read this list. CheckJudgement.cause still tells environment faults apart as before: the full-disk mark comes from fullShard(lines) on the raw log, not from the names. rerunnable still refuses a re-run whenever the named failures don't add up to the failure count, so fallout can't make a red shard re-runnable.

Tests (test/checks.test.ts, in the parser suite):

• A shard whose named failure is followed by (unnamed) fallout names only the named failure.
• A shard with nothing but fallout is still named by it.
• Three shards logged in the order 3, 1, 2, with the tally echo after them, are listed in the order 1, 3.

One existing test changed: test/shared-red.test.ts (#4582, "summary names the file and the hook") expected a hook timeout followed by two (unnamed) entries. Under this issue's rule that shard has a named failure, so it now expects only a.test.ts > beforeEach/afterEach (timeout) and a summary ending "and 1 more". That test recorded the old behaviour, which this issue replaces. A side effect: an afterAll that throws in the same shard, which shows up as (unnamed): error: …, is no longer named in the summary, only counted. Its line is still in the log. Renamed hook timeouts count as named failures, not fallout, because the issue asks only about (unnamed).

Not done: no change to how scripts/test-shards.ts writes the log. Its final tally was already in shard order; the summary was taking its names from the per-shard sections, which are in finish order.

Checks run: bun scripts/gate.ts, bun x tsc --noEmit and the 39 test files that pass (fail), (unnamed) or checkDetail through the parser all pass (497 tests). scripts/fast.ts picked 268 files, about 24 minutes, so I ran those 39 instead and left the full run to the forge's check.

To try it: bun test test/checks.test.ts test/shared-red.test.ts test/disk-full-shard.test.ts

🤖 Generated with Claude Code