Search perf traces

Capture harness for the gi-kg-viewer Search surface (Query Workspace, RFC-107). Two capture paths: UI (browser paint/interaction timings) and API (endpoint latency, no browser).

• UI: scripts/dev/capture-search-perf.{sh,mjs}
• API: scripts/dev/capture-search-api.{sh,py}
• Deep recipe: SEARCH_PERF_TRACE_RUNBOOK.md
• Raw artifacts: data/perf/traces/search/

See index.md for the framework, standard conditions (prod-v2, 1440×900 @ DPR-2, median-of-3), and the split rationale.

Run it

# UI scenarios (boots dedicated api + viewer, tears down):
scripts/dev/capture-search-perf.sh \
  --corpus .test_outputs/manual/prod-v2/corpus \
  --label <release>-search-ui \
  --output-dir data/perf/traces/search

# API-only latency:
scripts/dev/capture-search-api.sh \
  --corpus .test_outputs/manual/prod-v2/corpus \
  --queries tests/fixtures/viewer-validation-corpus/v3/search-queries.json \
  --label <release>-search-api \
  --output-dir data/perf/traces/search

Scenario catalog (UI)

Status reflects what capture-search-perf.mjs implements today. RFC-107 §P2 enumerated the full target set; Chunk 1 of the perf-resolidify pass fills the S2–S8 scenarios.

Scenario	Measures	Status
workspace-open (TTI)	page load → set corpus path → Search tab → #search-q visible	Implemented
filter-apply	top-k chip click → popover visible	Implemented
results-paint	submit query → first result card visible	Implemented
cmdk-open	/ → palette overlay visible (S3)	Implemented
operator-cluster	Cluster chip → cluster panel populated (S4)	Implemented
operator-compare	Compare run → 2-column packs rendered (S8)	Implemented

--query <q> overrides the results/operator query (default the economy, suited to the prod-v2 finance corpus). Every scenario degrades gracefully (records an error field, never aborts the run) so a report is always produced. Latest run: data/perf/traces/search/2.7.0.dev1-search-ui.ui.metrics.json.

Scenario catalog (API)

capture-search-api.py emits one aggregated *.api.metrics.json of per-query latency over the labelled query set (see the runbook). The committed S4-shell-fixture-baseline.api.metrics.json is the S4-shell baseline.

Reports

Per-release numbers + comparison tables live under reports/.