[kunkunblueberry]

Follow-up to #216 (already merged). Fixes the [P2] raised by @hsliuustc0106 on that PR: a required measurement missing from a sample that otherwise parsed never reached the failure decision.

What was wrong

build_report recorded a missing measurement in missing and then ignored it. notify only looked at regressions, suite, and ui_required and ui_unavailable (which is ui is None), so a sample that parsed but carried no measurement stayed green.

Reproduced against 2f35c78e by calling build_report directly with storage samples that are all ok: true:

input	missing	notify	summary tail
ui = {"drag": {}, "render": {}}, ui_required=True	both ui.drag.*.frameP95	False	within threshold
storage sample with cold-open.data.open removed	storage.coldOpen.open	False	... 400.0ms/FAILED ... within threshold

The second row is the worst shape: the summary itself printed FAILED for that clause while the lane exited 0.

What changed

• required_metric_ids() makes the requirement explicit: storage metrics always, drag metrics only while --ui-required is passed. An advisory run whose UI lane never ran keeps its current behaviour.
• build_report computes incomplete = [id for id in missing if id in required], feeds it to format_summary, and adds it to notify. The artifact carries it as incomplete ([] in failure_report, so the document shape stays stable).
• The one-line summary names the gap as INCOMPLETE <ids> and no longer appends within threshold when one exists.

Docs and workflow comments that stated the failure condition (docs/development/validation.md, .github/workflows/nightly.yml) now include the incomplete case.

Validation

python -m pytest tests/test_daily_benchmark.py -q --noconftest
16 passed (13 existing + 3 new)

The three new tests are the reproductions above plus the guard that a partial UI sample without --ui-required still stays advisory:

• test_empty_ui_sample_is_incomplete_when_the_lane_requires_it
• test_storage_sample_missing_a_required_field_is_incomplete
• test_partial_ui_sample_stays_advisory_while_the_lane_is_optional

yaml.safe_load on the edited workflow parses, and the daily-benchmark job keeps its 8 steps with the renamed failure step.

Not covered here

test_storage_bench_emits_the_fields_the_suite_tracks runs the real storage bench, but no new test executes the whole workflow; the lane's own nightly run is the end-to-end check.

Refs: #216