vvsandClaude Opus 5
99be33a632
X5: dissimilar+slope with A reaches 2859+2545 = 5404 over both sets, the best total this line has measured (was 5511; Chandrupatla 5228) - but again by trading the canonical set (2726 -> 2859) for the wide brackets (2785 -> 2545). The one-step replay (APS no harm) did not carry over: its bracket-halving metric rewards overshooting the root, which on a simple root costs superlinear convergence in the full run - the declared one-step limitation, now measured. Among the nine functions A helps only the tent (6026 -> 5892) and slightly ratio_soft. Also: pin 'strictly fewer evaluations on every flat problem' was wrong for 6(x-2)^5 on [0,10] - both runs are budget-bound (90 of the guaranteed 91); corrected to the measured claim. verify_article 286/286, 836 tests pass twice. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
42 lines
2.8 KiB
Plaintext
42 lines
2.8 KiB
Plaintext
tent, three-point, pairs (without A -> with A): APS | CHA | both | flat fun3-6 sum
|
|
defect credit 3545->4134 2949->2926 6494->7060 1044->993 solved 193 over 0
|
|
plain credit 3568->4130 3020->2986 6588->7116 1044->988 solved 193 over 0
|
|
dissim+slope credit 3857->4223 3002->2947 6859->7170 1043->1002 solved 193 over 0
|
|
slope credit 3507->4121 2997->3007 6504->7128 1051->1007 solved 193 over 0
|
|
shape+slope credit 3755->4169 3021->3059 6776->7228 1064->1064 solved 193 over 0
|
|
defect none 2523->3180 4198->3508 6721->6688 2806->2081 solved 188 over 1
|
|
plain none 2806->2967 4104->3471 6910->6438 2707->2070 solved 188 over 0
|
|
dissim+slope none 2726->2859 2785->2545 5511->5404 1376->1157 solved 193 over 0
|
|
slope none 2848->2971 3075->2735 5923->5706 1710->1331 solved 193 over 0
|
|
shape+slope none 2675->2930 5443->5452 8118->8382 4000->4000 solved 173 over 0
|
|
defect project 2523->3180 3210->2954 5733->6134 1818->1527 solved 193 over 0
|
|
plain project 2806->2967 3220->2925 6026->5892 1823->1524 solved 193 over 0
|
|
dissim+slope project 2726->2859 2785->2545 5511->5404 1376->1157 solved 193 over 0
|
|
slope project 2848->2971 3074->2735 5922->5706 1709->1331 solved 193 over 0
|
|
shape+slope project 2675->2930 3498->3506 6173->6436 2055->2054 solved 193 over 0
|
|
defect safe 3131->3207 3368->3386 6499->6593 1943->1929 solved 193 over 0
|
|
plain safe 3036->3171 3363->3389 6399->6560 1950->1931 solved 193 over 0
|
|
dissim+slope safe 3123->3170 2683->2710 5806->5880 1228->1248 solved 193 over 0
|
|
slope safe 3020->3187 2903->2895 5923->6082 1484->1469 solved 193 over 0
|
|
shape+slope safe 2527->2986 3108->3223 5635->6209 1629->1687 solved 193 over 0
|
|
|
|
VERDICTS for the proposed rule (tent, 3-point, projection)
|
|
X1 CHA < 3220: 2925 -> PASS
|
|
X2 APS <= 2806: 2967 -> FAIL
|
|
X3 guarantee family with A: 0 violations over 39 cells; cells solving < 193: 0 -> PASS
|
|
X4 flat fun3-6 sum: 1823 -> 1524 -> PASS
|
|
X5 best pair with A over both sets: pure+dissim+ext+slope.ratio.derived3.project 2859+2545=5404 (reference 5511, Chandrupatla 5228)
|
|
proposed + A vs Chandrupatla, APS: 42/62/49 sign p=0.53 totals 2967 vs 2787
|
|
proposed + A vs Chandrupatla, CHA: 11/1/28 sign p=0.0095 totals 2925 vs 2441
|
|
|
|
nine functions, three-point, projection, no readings (without A -> with A), both sets:
|
|
cbrt 17993 -> 18181
|
|
cosine 19784 -> 19753
|
|
exp_abs 6652 -> 6869
|
|
exp_sq 19786 -> 19780
|
|
rat_sq 19793 -> 19796
|
|
ratio 6026 -> 5892
|
|
ratio_soft 9352 -> 9285
|
|
rational 6823 -> 7483
|
|
sqrt 13381 -> 13633
|