Thibaut Forest Claude Fable 5 commited on
Commit
a94f6b9
·
1 Parent(s): 4e17cd1

Phase 47 renumber: blind-champion adjudication (was duplicate Phase 46); register blind4/blind3 cv variants

Browse files

The blind4 writeup was swept into 4e17cd1 with a colliding Phase 46 header.
Renumbered to 47 per cross-session coordination. cv.py gains the rejected
blind4/blind3 variants with the verdict inline so the port cannot be
re-attempted by inspection.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

Files changed (2) hide show
  1. EXPERIMENTS.md +4 -1
  2. batteryswap_example/cv.py +15 -1
EXPERIMENTS.md CHANGED
@@ -2369,7 +2369,10 @@ forward-reading of the split file that phase 43(a) verified and declined.
2369
 
2370
  Standing champion: **3c7b883 at 1661.45**. Deadline 2026-08-23.
2371
 
2372
- ## Phase 46 - the blind champion itself, as a whole: REJECTED (lineage fully adjudicated)
 
 
 
2373
 
2374
  Question (user, 2026-08-22): the blind-repo multi-agent loop's shared champion
2375
  (`9a969a6a7dfe.py`, blind full-40 1566.7) - reusable here? The headline number
 
2369
 
2370
  Standing champion: **3c7b883 at 1661.45**. Deadline 2026-08-23.
2371
 
2372
+ ## Phase 47 - the blind champion itself, as a whole: REJECTED (lineage fully adjudicated)
2373
+
2374
+ (Numbered out of order: written before the phase-46 commit landed from the
2375
+ parallel session; the work predates it, the number does not.)
2376
 
2377
  Question (user, 2026-08-22): the blind-repo multi-agent loop's shared champion
2378
  (`9a969a6a7dfe.py`, blind full-40 1566.7) - reusable here? The headline number
batteryswap_example/cv.py CHANGED
@@ -327,9 +327,23 @@ def _v_last_only(t):
327
  return ['v_last']
328
 
329
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
330
  VARIANTS = {'base': _base, 'zone': _zone, 'vcorr': _vcorr, 'vrel': _vrel,
331
  'vrel_only': _vrel_only, 'vmin7_only': _vmin7_only,
332
- 'v_last_only': _v_last_only}
333
 
334
  ABLATION = ('base', 'vrel_only', 'vmin7_only', 'vrel', 'v_last_only')
335
 
 
327
  return ['v_last']
328
 
329
 
330
+ def _blind4(t):
331
+ # The blind-repo champion's EXACT feature set as a REPLACEMENT ranker
332
+ # (9a969a6a7dfe: last_voltage/vmin7/n_obs/vrel -> our columns). Phase 47:
333
+ # real head-ordering gain at fixed N (static N=8 +27 [+7,+47]) but the
334
+ # planner converts it to ZERO extra catches (290 vs 291 count-matched)
335
+ # and +79/origin cost via cruder day placement. REJECTED. blind3 shows
336
+ # vrel is not load-bearing - the effect is the small set itself.
337
+ return ['v_last', 'v_min_7', 'n_30', 'vrel']
338
+
339
+
340
+ def _blind3(t):
341
+ return ['v_last', 'v_min_7', 'n_30']
342
+
343
+
344
  VARIANTS = {'base': _base, 'zone': _zone, 'vcorr': _vcorr, 'vrel': _vrel,
345
  'vrel_only': _vrel_only, 'vmin7_only': _vmin7_only,
346
+ 'v_last_only': _v_last_only, 'blind4': _blind4, 'blind3': _blind3}
347
 
348
  ABLATION = ('base', 'vrel_only', 'vmin7_only', 'vrel', 'v_last_only')
349