Surface perf traces (Library · Digest · entity-load)¶
Capture harness for the basic-but-important viewer surfaces outside search and graph — the ones a user hits constantly and that must not regress silently.
- UI:
scripts/dev/capture-surface-perf.{sh,mjs} - Raw artifacts:
data/perf/traces/surfaces/
See index.md for the framework, standard conditions, and split rationale.
Run it¶
scripts/dev/capture-surface-perf.sh \
--corpus .test_outputs/manual/prod-v2/corpus \
--label <release>-surfaces \
--output-dir data/perf/traces/surfaces
Scenario catalog¶
| Scenario | Measures | Status |
|---|---|---|
library-load |
Library tab click → library-root visible |
Implemented |
digest-load |
Digest tab click → digest-root visible |
Implemented |
entity-load |
Library → first episode row click → episode-detail-rail visible |
Implemented |
What the numbers mean: the corpus envelope is fetched on landing, so these
measure tab-switch / interaction → container-visible responsiveness (not a
cold data fetch). That's the right signal for these surfaces — the regression we
care about is "a change made switching to Library / opening an episode janky."
Every scenario degrades gracefully (records an error field, never aborts).
These are deliberately "boring" surfaces: the point is a stable baseline so a
future change that quietly doubles Library responsiveness is caught at the next
release capture, not by a user. Latest run:
data/perf/traces/surfaces/2.7.0.dev1-surfaces.ui.metrics.json.
Reports¶
Per-release numbers live under reports/.