Metric defect found by B's first run (mean = inf): an exact hit made the new width 0 and the floor (ulp of the root, 5e-324 when the root is 0) overflowed. Fixed to XTOL - an exact hit ends the run, so it is worth the halvings still needed to reach xtol - identical for every candidate; A re-run with it.

A with the fixed metric: still PASSES A1 and A2 in both configurations - flat 0.26 -> 2.82 (proposed) and 0.35 -> 0.77 (reading 3), canonical 6.32 -> 8.49 and 9.00 -> 10.44 - but a correction to the first report: for reading 3 on the flat families step A (0.77) beats our step and NOT bisection (0.99); the first run said 1.16. B (outcome memory) FAILS its falsifier: flat 0.31 -> 0.97 and 0.58 -> 0.82 (B1 holds) but canonical 4.74 -> 2.49 and 5.52 -> 2.12 (B2 fails), bisecting on 66% and 53% of canonical states. Mechanism: the known trap - a good step can collapse the error without moving the bracket, so an outcome measured in bracket shrink condemns it; the same trap that sank the bracket-progress gate (4449) and the bracket guard (7260). Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-09-21 11:49:39 +01:00
parent 2b0ee228b4
commit 2498c66358
3 changed files with 52 additions and 1 deletions
@@ -0,0 +1,14 @@
+
+== proposed (projection)
+   group            states  B bisects    ours       B   (bracket halvings per eval, all states with a decision)
+   APS                1258      65.6%     inf     inf
+   CHA fun1             46      60.9%    5.44    2.17
+   CHA flat fun3-6    1623      83.6%    0.32    0.98
+   CHA other           165      73.3%    4.88    2.03
+
+== reading (3), as for m>=2
+   group            states  B bisects    ours       B   (bracket halvings per eval, all states with a decision)
+   APS                 958      52.6%    7.46    2.69
+   CHA fun1             32      43.8%    5.01    1.52
+   CHA flat fun3-6     603      48.3%    0.59    0.83
+   CHA other           120      46.7%    6.59    2.72