## Summary - rebalance Browser tests with explicit class-filter shards - build PR browser assets inside shards, removing the build-assets dependency gate - run Browser shards with Pest parallelism () ## Autoresearch metric - baseline modeled PR critical path: 406.50s - final modeled PR critical path: 234.27s - modeled improvement: 172.23s (42.4%) ## Verification - ran METRIC ci_total_s=234.270 METRIC actual_recent_pr_total_s=422.000 METRIC build_assets_s=0.000 METRIC tests_s=170.000 METRIC browser_matrix_s=232.270 METRIC linter_s=60.000 METRIC static_analysis_s=26.500 METRIC performance_tests_s=63.000 METRIC job_count=13.000 METRIC browser_shards=6.000 for each experiment - coverage guard in autoresearch script checks Browser filters cover all recent Browser classes exactly once ## Notes - wall-clock faster; runner minutes likely higher due extra shards and duplicate asset builds - real CI should validate Browser parallelism flake risk |
||
|---|---|---|
| .. | ||
| shards.json | ||