vvs/sim-numerics
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

PREREG_defect_contra_flat outcome (smoke, all 199 problems). Goal (below Chandrupatla on both) NOT reached by any cell: best CHA 2537 vs 2441. Best total: defect+flatanalogmem APS 2530 / CHA 2561 = 5091 < Chandrupatla 5228 (first time), C1-C3 hold; ablation: defect earns APS +284 (costs CHA 9), remembered analogy earns CHA +649 (costs APS 7). contra earned without defect (APS 2806 -> 2785, CHA 3220 -> 3198; with the remembered analogy 2792 / 2537, the best CHA), redundant with defect (defect+contra = defect). fun9: 105 -> 81 (Chandrupatla 55). Remaining CHA gap 120: flat +88, fun1 +20, fun9 +26. Server regression x2: 1116 passed, 1 failed = the paper-number check.

Browse Source 
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
vvs
2026-09-21 15:57:15 +01:00
co-authored by Claude Opus 5
parent 3c92d610d6
commit 3d9287ec2c
6 changed files with 205 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
results_analyze_defcontra_smoke.txt
+38
View File
@@ -0,0 +1,38 @@
== proposed rule: baseline APS 2806 CHA 3220
prestep APS CHA both C1 C2 C3 goal vs Chandrupatla APS / CHA (w/t/l)
none 2806 3220 6026 ✗ ✓ ✓ no 87/24/42 p=9.2e-05 | 9/5/26
defect 2523 3210 5733 ✓ ✓ ✓ no 86/21/46 p=0.00063 | 7/4/29
contra 2785 3198 5983 ✓ ✓ ✓ no 90/23/40 p=1.4e-05 | 9/5/26
flatanalog 2807 2635 5442 ✓ ✗ ✓ no 87/24/42 p=9.2e-05 | 9/6/25
flatanalogmem 2814 2552 5366 ✓ ✗ ✓ no 87/24/42 p=9.2e-05 | 9/7/24
defect+flatanalog 2526 2624 5150 ✓ ✓ ✓ no 86/20/47 p=0.00091 | 8/4/28
contra+flatanalog 2785 2620 5405 ✓ ✓ ✓ no 90/23/40 p=1.4e-05 | 8/6/26
defect+contra 2523 3210 5733 ✓ ✓ ✓ no 86/21/46 p=0.00063 | 7/4/29
defect+flatanalogmem 2530 2561 5091 ✓ ✓ ✓ no 86/20/47 p=0.00091 | 8/5/27
defect+contra+flatanalog 2526 2624 5150 ✓ ✓ ✓ no 86/20/47 p=0.00091 | 8/4/28
contra+flatanalogmem 2792 2537 5329 ✓ ✓ ✓ no 90/23/40 p=1.4e-05 | 8/7/25
defect+contra+flatanalogmem 2530 2561 5091 ✓ ✓ ✓ no 86/20/47 p=0.00091 | 8/5/27
best accepted cell: defect+flatanalogmem (both 5091)
C4 ablation - remove defect : APS +284 CHA -9 (positive = the component earns it)
C4 ablation - remove flatanalogmem : APS -7 CHA +649 (positive = the component earns it)
CHA per family (ours/Chandrupatla): fun1 123/103 fun2 144/152 fun3 304/305 fun4 315/297 fun5 257/233 fun6 280/233 fun7 1000/1000 fun8 57/63 fun9 81/55
== proposed rule + G2: baseline APS 2782 CHA 3169
prestep APS CHA both C1 C2 C3 goal vs Chandrupatla APS / CHA (w/t/l)
flatanalog 2780 2625 5405 ✓ ✓ ✓ no 95/20/38 p=8.2e-07 | 9/6/25
flatanalogmem 2787 2548 5335 ✓ ✗ ✓ no 95/20/38 p=8.2e-07 | 9/7/24
defect+flatanalog 2583 2622 5205 ✓ ✓ ✓ no 85/20/48 p=0.0017 | 8/3/29
defect+flatanalogmem 2587 2564 5151 ✓ ✓ ✓ no 85/20/48 p=0.0017 | 8/4/28
contra+flatanalog 2766 2625 5391 ✓ ✓ ✓ no 98/18/37 p=1.5e-07 | 8/6/26
defect+contra+flatanalog 2583 2622 5205 ✓ ✓ ✓ no 85/20/48 p=0.0017 | 8/3/29
contra+flatanalogmem 2773 2548 5321 ✓ ✓ ✓ no 98/18/37 p=1.5e-07 | 8/7/25
defect+contra+flatanalogmem 2587 2564 5151 ✓ ✓ ✓ no 85/20/48 p=0.0017 | 8/4/28
defect 2580 3182 5762 ✗ ✓ ✓ no 85/21/47 p=0.0012 | 8/2/30
contra 2766 3173 5939 ✗ ✓ ✓ no 98/18/37 p=1.5e-07 | 9/3/28
defect+contra 2580 3182 5762 ✗ ✓ ✓ no 85/21/47 p=0.0012 | 8/2/30
none 2782 3169 5951 ✗ ✓ ✓ no 95/20/38 p=8.2e-07 | 10/3/27
best accepted cell: defect+flatanalogmem (both 5151)
C4 ablation - remove defect : APS +200 CHA -16 (positive = the component earns it)
C4 ablation - remove flatanalogmem : APS -7 CHA +618 (positive = the component earns it)
CHA per family (ours/Chandrupatla): fun1 123/103 fun2 145/152 fun3 306/305 fun4 315/297 fun5 257/233 fun6 280/233 fun7 1000/1000 fun8 57/63 fun9 81/55