Skip to content

StreamVault is a TV-first free IPTV player for Android TV built with Kotlin, Jetpack Compose, Room, Hilt, and Media3.

docs(catalog): record dashboard benchmark evidence

David Nashashfe402aa36e3f
diff --git a/docs/COMPOSE_REDUCTION_AND_UI_ARCHITECTURE_PLAN.md b/docs/COMPOSE_REDUCTION_AND_UI_ARCHITECTURE_PLAN.md
index 7f71ad1c..7acaa803 100644
--- a/docs/COMPOSE_REDUCTION_AND_UI_ARCHITECTURE_PLAN.md
+++ b/docs/COMPOSE_REDUCTION_AND_UI_ARCHITECTURE_PLAN.md
@@ -980,9 +980,10 @@ paths, plus reversible quick-filter save/reload/remove persistence. The full
regenerated with its connected assertion passing. Catalog is now structurally
extracted under `:feature:catalog`; its execution report records passing
connected goldens, a partial seeded Home/Live/Movies/Series/Search smoke pass,
- unavailable full catalog journeys, a Dashboard benchmark attempt blocked by
- missing render-thread trace slices, an open paired benchmark gate, refreshed
- release-like profiles, and current app lint baseline failure. System remains
+ unavailable full catalog journeys, a successful diagnostic Dashboard benchmark
+ rerun after populating a scrollable seeded Home shelf, an open paired
+ benchmark gate, refreshed release-like profiles, and current app lint baseline
+ failure. System remains
unstarted; Playback, Provider, and Settings gates remain governed by their
separate reports.
diff --git a/docs/COMPOSE_REDUCTION_PHASE5_CATALOG_REPORT.md b/docs/COMPOSE_REDUCTION_PHASE5_CATALOG_REPORT.md
index 5bba1196..b47b8a2c 100644
--- a/docs/COMPOSE_REDUCTION_PHASE5_CATALOG_REPORT.md
+++ b/docs/COMPOSE_REDUCTION_PHASE5_CATALOG_REPORT.md
@@ -115,10 +115,17 @@ An actual Catalog-only edit followed by `:app:assembleDebug` executed the
Catalog Kotlin task; sibling feature Kotlin tasks were `UP-TO-DATE` and did
not execute. A clean debug assembly passed in 80.7s, warm debug assembly in
14.9s, and Beta/Release packaging passed in 332.3s. These are not a
-cache-equivalent five-run clean-build comparison. The focused Dashboard
-`dashboardVerticalScroll` macrobenchmark was attempted on the same emulator,
-but failed before metrics with `Observed no renderthread slices in trace` from
-`FrameTimingQuery`; no after P50/P90/P99 values were recorded.
+cache-equivalent five-run clean-build comparison. A focused Dashboard
+`dashboardVerticalScroll` rerun on the same emulator passed 1/1 with five warm
+iterations after public-M3U live-channel activity populated a scrollable Home
+recent-channel shelf. It emitted `frameCount` min/median/max 88/91/98,
+`frameDurationCpuMs` P50/P90/P95/P99 60.0/70.7/78.7/92.7, and
+`frameOverrunMs` P50/P90/P95/P99 63.1/78.9/89.7/106.1. The earlier empty Home
+fixture had no scrollable content and failed before metrics with
+`Observed no renderthread slices in trace` from `FrameTimingQuery`. This rerun
+is diagnostic after-run evidence only: no cache-equivalent pre-extraction run
+was captured, and the public M3U fixture still lacks VOD/series content, so the
+formal paired performance gate remains open.
## Profiles and known failures
@@ -150,7 +157,7 @@ execution-time project access.
1. Add a seeded production-activity Catalog fixture and execute the documented
journeys, including Favorites and accessibility variants.
-2. Re-run the Dashboard macrobenchmark on a device/trace configuration that
- emits render-thread slices, then complete the cache-equivalent paired
- performance comparison. Keep app lint remediation and all other Phase 5
- feature gates tracked separately.
+2. Capture a cache-equivalent pre-extraction run and pair it with the
+ successful Dashboard benchmark evidence, using the same seeded content,
+ device, iteration count, and compilation mode. Keep app lint remediation and
+ all other Phase 5 feature gates tracked separately.
diff --git a/docs/COMPOSE_REDUCTION_PHASE5_ROADMAP.md b/docs/COMPOSE_REDUCTION_PHASE5_ROADMAP.md
index db43893a..8c7b83a2 100644
--- a/docs/COMPOSE_REDUCTION_PHASE5_ROADMAP.md
+++ b/docs/COMPOSE_REDUCTION_PHASE5_ROADMAP.md
@@ -116,8 +116,12 @@ Catalog extraction is structurally implemented with passing connected golden
checks and current-checkout build-isolation samples. A partial seeded
Home/Live/Movies/Series/Search smoke pass is recorded; full VOD/detail,
Favorites, accessibility, and production journey coverage remains unavailable.
-The Dashboard macrobenchmark was attempted but the emulator emitted no
-render-thread slices, so no frame metrics were produced; the paired comparison
-and repository-wide app lint remain open.
+A focused Dashboard macrobenchmark rerun passed 1/1 with five warm iterations
+once public-M3U live-channel activity populated a scrollable Home shelf
+(`frameCount` min/median/max 88/91/98; `frameOverrunMs` P50/P90/P95/P99
+63.1/78.9/89.7/106.1). The earlier empty Home fixture emitted no render-thread
+slices; the rerun is diagnostic after-run evidence, not a cache-equivalent
+paired comparison. The paired comparison and repository-wide app lint remain
+open.
The release-like profile workflow now passes with feature Catalog descriptors.
System remains unstarted.
diff --git a/docs/superpowers/plans/2026-09-02-catalog-feature-extraction.md b/docs/superpowers/plans/2026-09-02-catalog-feature-extraction.md
index 7d6dc64c..654a1b66 100644
--- a/docs/superpowers/plans/2026-09-02-catalog-feature-extraction.md
+++ b/docs/superpowers/plans/2026-09-02-catalog-feature-extraction.md
@@ -29,9 +29,11 @@
Tasks 0–12 are implemented and their evidence is recorded under
`validation/phase5_catalog/`. Task 13 structural/build-isolation and profile
work is also recorded there. The available seeded Home/Live/Movies/Series/
-Search smoke pass is documented; full seeded journeys, the Dashboard
-macrobenchmark (attempted but blocked by missing render-thread trace slices),
-and repository-wide app lint remain open or unavailable as documented in
+Search smoke pass is documented; full seeded journeys and repository-wide app
+lint remain open or unavailable. A focused Dashboard macrobenchmark rerun now
+passes after public-M3U live-channel activity populates a scrollable seeded Home
+shelf, but it is diagnostic after-run evidence rather than the required
+cache-equivalent paired comparison. These limitations are documented in
`docs/COMPOSE_REDUCTION_PHASE5_CATALOG_REPORT.md`. Do not interpret this
status as completion of Phase 5.
diff --git a/validation/phase5_catalog/performance-after.md b/validation/phase5_catalog/performance-after.md
index 65b15632..00d84df6 100644
--- a/validation/phase5_catalog/performance-after.md
+++ b/validation/phase5_catalog/performance-after.md
@@ -55,19 +55,32 @@ Additional guardrails:
These are current-checkout samples, not a cache-equivalent five-run clean-build
comparison.
-## Dashboard macrobenchmark attempt
+## Dashboard macrobenchmark diagnostic run
-The focused `dashboardVerticalScroll` benchmark was attempted against the same
-API 36 TV emulator after installing the benchmark and release artifacts:
+The focused `dashboardVerticalScroll` benchmark was rerun against the same API
+36 TV emulator after installing the benchmark and seeded debug artifacts:
```text
-./gradlew.bat :benchmark:connectedBenchmarkBenchmarkAndroidTest
- -Pandroid.testInstrumentationRunnerArguments.class=
- com.streamvault.benchmark.StreamVaultMacrobenchmark#dashboardVerticalScroll
+./gradlew.bat :benchmark:connectedBenchmarkBenchmarkAndroidTest `
+ '-Pandroid.testInstrumentationRunnerArguments.class=com.streamvault.benchmark.StreamVaultMacrobenchmark#dashboardVerticalScroll' `
+ --no-daemon --console=plain --warning-mode=none
```
-The run reached the device but failed before emitting frame metrics with
-`IllegalStateException: Observed no renderthread slices in trace` from
-`FrameTimingQuery`. No P50/P90/P99 after values were recorded, so the formal
-cache-equivalent paired performance gate remains open pending a device/trace
-configuration that produces render-thread slices.
+Before this run, the seeded debug app had accumulated recent-channel data by
+opening public-M3U live channels. That made the Home `LazyColumn` scrollable;
+the earlier no-data fixture had no scrollable content and produced
+`IllegalStateException: Observed no renderthread slices in trace` before
+metrics. The rerun passed 1/1 test with five warm iterations and emitted:
+
+| Metric | Result |
+|---|---|
+| `frameCount` | min 88, median 91, max 98 |
+| `frameDurationCpuMs` | P50 60.0, P90 70.7, P95 78.7, P99 92.7 |
+| `frameOverrunMs` | P50 63.1, P90 78.9, P95 89.7, P99 106.1 |
+
+The five Perfetto traces and the raw metric message are retained in the
+Gradle-connected-test output directory. This is actionable after-run evidence,
+but not a cache-equivalent paired before/after comparison: no equivalent
+pre-extraction run was captured in this continuation, and the public M3U
+fixture still lacks VOD/series content. Keep the formal paired performance gate
+open.
diff --git a/validation/phase5_catalog/profile-validation.md b/validation/phase5_catalog/profile-validation.md
index 963b54e1..c2b39578 100644
--- a/validation/phase5_catalog/profile-validation.md
+++ b/validation/phase5_catalog/profile-validation.md
@@ -20,5 +20,8 @@ Home/Live journeys; the public M3U fixture does not provide VOD/movie/series
content, so it is profile evidence rather than full Catalog journey coverage.
The current packaging guardrail (`:app:assembleBeta` and
-`:app:assembleRelease`) passed after generation. The Dashboard macrobenchmark
-remains open because its configured test is skipped.
+`:app:assembleRelease`) passed after generation. The configured profile suite
+still skips its eight ordinary macrobenchmark cases. Separately, the focused
+Dashboard `dashboardVerticalScroll` diagnostic rerun passed after seeded
+recent-channel data made Home scrollable; it does not close the required
+cache-equivalent paired performance comparison.