1.9 KiB
B1 labelled recovery and scoring extension — final result
Of1167 geometry candidates,29 v4 parts never have a requested test prediction in the fixed random splits and remain unscored. Canonical1138 requested protocol/part scores and 12772 field/topology rows complete. Original20 models and predictions retained; extension reused1109 completed scores and ran29 unfinished/cut scores plus summary, all30 guards PASS. No refit, new prediction or changed scoring function.
Grouped100%:552 parts/204 components. H1−A field relative L2 = −0.000637370497,95%CI[−0.001141815998,−0.000163126009]; means H1 .070193170323 versus A .070830540821. This is a small observed improvement, about0.063737 percentage points; it does not clear the frozen0.1 percentage-point margin. Last-location Δ=.006509692063,CI[−.037840008549,.049622005664]; Kendall Δ=−.000130998912,CI[−.000646520891,.000334445350]. No secondary improvement gate passes. Frozen tree decision PARK/inconclusive, not zero effect or broad representation failure.
Cumulative postfit26087.366284311/61200CPU-s; total32502.931080962/72000. Original25200 stop remains immutable; settled prior25377.194583929 includes177.194584 overshoot, not the worker/guard display22020.4754. Extension new710.171700382CPU-s. Bootstrap9 original-function calls=.231703283CPU-s.
Canonical completed field+input CPU144.277951596; topology incremental CPU11281.776495508. These timers exclude original duplicate scoring and interrupted attempts; the difference from total postfit is not assignable solely to topology. See cost-breakdown.json for boundary.
Final audit checked16884 reused file hashes, all pinned inputs/code and1138 complete pairs. No metrics recomputed. Original NA, original failed five predictions, v2 interrupted score receipts and all costs remain. v4 and topology are descriptive only. H2 is independently authorized; this result does not reopen O1/type/connectivity or authorize another fit.