Files
MeltMind/docs/phase2-b13-results-2026-11-17.md
T

16 lines
2.0 KiB
Markdown
Raw Blame History

This file contains ambiguous Unicode characters
This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.
# B13 fixed one-fold gate — terminal result
Decision **EXPAND_THIS_FIXED_RECIPE**. Full learning320 parts/132 components, validation111 parts/35 components. One model; all431 geometry attempts and train/predict/score guards PASS. Total1761.248712587/7200CPU-s.
relative_l2: new mean0.0632306317402, A0.064863959003; new−A-0.00163332726276, component-paired95%CI[-0.0033260470954,-0.000114453529849],35 components,2000 replicates,undefined0.
last_location: new mean0.377147230826, A0.294753956234; new−A0.0823932745924, component-paired95%CI[-0.0360340511063,0.235476347466],35 components,2000 replicates,undefined0.
Geometry coverage: {"learning": {"components": 132, "complete_components": 107, "fraction": 0.8106060606060606, "valid_parts": 288, "total_parts": 320}, "validation": {"components": 35, "complete_components": 32, "fraction": 0.9142857142857143, "valid_parts": 107, "total_parts": 111}}.
Validation invalid-G/geometry parts: 4; IDs retained in original score result. Undefined field metrics: 0.
G-valid learning288/320 (32 unavailable), validation107/111 (4 unavailable). Component-complete coverage107/132=.810606 and32/35=.914286 passes both80% gates. All320 learning and111 validation parts remain in the fit/evaluation; unavailable G contributes nativeNaN in the two new columns, original19 finite inputs preserved. No hard cases dropped. Valid-only107 subset is descriptive and does not replace full111 gate.
Field upper bound−.00011445353<0 triggers the first OR branch. Last-location interval crosses0; do not claim improvement there. This supports this fixed generic-region context recipe only, not type benefit, physical conservation, or five-fold confirmation. B13-E1/E2 remain plans, no launch. The frozen implementation did not compute Kendall tau; do not invent that diagnostic.
Audit verified all feature/result hashes,111 prediction files, checkpoint and source pins without recomputation. Full raw artifacts remain available; this report does not claim independent team review already completed.