Skip to content

feat(profiling): P5-① e2e hybrid recall@10 (LOC-70)#76

Merged
dev07060 merged 11 commits into
mainfrom
feat/loc-70-recall
Jun 1, 2026
Merged

feat(profiling): P5-① e2e hybrid recall@10 (LOC-70)#76
dev07060 merged 11 commits into
mainfrom
feat/loc-70-recall

Conversation

@dev07060

@dev07060 dev07060 commented Jun 1, 2026

Copy link
Copy Markdown
Owner

Summary

  • Adds Dart-side f32 ground-truth recall helpers and read-only SQLite corpus loading for the example profiler.
  • Adds an iPhone profile recall harness comparing frozen-query GT against vector-only and shipped hybrid search.
  • Records the P5-① result in docs/perf/ondevice-query-profiler/PR-P5-1.html and updates the profiler README status.

Result

  • iPhone profile run passed with --no-dds.
  • mean recall_vectoronly@10 = 1.00
  • mean recall_hybrid@10 = 0.08
  • Verdict: vector-only quality gate passes; hybrid value is a BM25/RRF reorder diagnostic against pure-vector GT, not an HNSW failure.

Verification

  • flutter test test/profiling/recall_math_test.dart test/profiling/recall_db_test.dart test/profiling/recall_report_test.dart
  • flutter test -d macos integration_test/query_recall_measure_test.dart
  • flutter drive --driver=test_driver/integration_test.dart --target=integration_test/query_recall_measure_test.dart --profile --no-keep-app-running --no-dds --device-timeout=60 -d 00008110-001524992E38801E

@linear-code

linear-code Bot commented Jun 1, 2026

Copy link
Copy Markdown

LOC-70

@dev07060 dev07060 merged commit 64ef3c2 into main Jun 1, 2026
6 checks passed
@dev07060 dev07060 deleted the feat/loc-70-recall branch June 1, 2026 08:39
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant