From 469d6ce606a2a00dd6695ce2bd7c7851c0088799 Mon Sep 17 00:00:00 2001 From: Yuren Hao Date: Wed, 29 Jul 2026 12:16:02 -0500 Subject: =?UTF-8?q?RESULT=2072:=20=E5=89=82=E9=87=8F=E6=9B=B2=E7=BA=BF?= =?UTF-8?q?=E9=80=9A=E9=A1=B6=20=E2=80=94=20dg128=E5=85=B3=E9=97=AD97%(?= =?UTF-8?q?=E8=B7=9Dall-BP=E6=9D=860.0015),=E7=89=A9=E7=90=86=E6=B3=84?= =?UTF-8?q?=E6=BC=8F=E5=AE=8C=E5=85=A8=E5=8F=AF=E8=B5=8E=E5=9B=9E;=20?= =?UTF-8?q?=E5=AF=B9=E6=95=B0=E5=BE=8B=E5=85=A8=E7=A8=8B=E6=88=90=E7=AB=8B?= =?UTF-8?q?34=E2=86=9297%;=20dgrand=E8=B4=A5=E4=BA=8E=E5=9B=BA=E5=AE=9A?= =?UTF-8?q?=E6=A1=A3;=20=E6=BB=A1=E5=89=82=E9=87=8F=E9=AA=8C=E8=AF=81fw135?= =?UTF-8?q?m=5Fdg128=E4=B8=89=E5=8D=A1=E5=8F=91=E5=B0=84(~22h,=E5=88=A4?= =?UTF-8?q?=E6=8D=AE=3D=E7=BB=88=E7=82=B9=E5=9B=9E72M=20gap=E5=B8=A6);=20?= =?UTF-8?q?=E9=81=97=E7=95=99=E9=97=AE=E9=A2=98=3D=E5=89=82=E9=87=8F-?= =?UTF-8?q?=E5=AE=BD=E5=BA=A6=E6=A0=87=E5=BA=A6(270M=E7=AD=94)?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Co-Authored-By: Claude Fable 5 Claude-Session: https://claude.ai/code/session_014FAPDWQ49M5Ye3NpTndTpn --- docs/campaign/CASCADE_ABLATION_PLAN.md | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) (limited to 'docs') diff --git a/docs/campaign/CASCADE_ABLATION_PLAN.md b/docs/campaign/CASCADE_ABLATION_PLAN.md index f919db1..f3678b3 100644 --- a/docs/campaign/CASCADE_ABLATION_PLAN.md +++ b/docs/campaign/CASCADE_ABLATION_PLAN.md @@ -568,6 +568,22 @@ direction), not training-under-fault; wave-2 = co-training with faults injected +### RESULT 72 (2026-07-29): THE CURVE GOES ALL THE WAY — dg128 closes 97% (tail within +0.0015 of the all-BP bar); the physics leak is FULLY REDEEMABLE by displacement amplification. +FULL-DOSE VALIDATION LAUNCHED. +Battery 8: dg64 92%, dg128 97% (log-law held the whole way: 34/50/63/75/92/97 at x4..x128), +dgrand64 64% (spread-spectrum loses to fixed dose — the value lives in the high octaves; +log-uniform wastes steps low). dg128 mechanics: r18/6k mild, no skips, drift 0.009; gn 5.07 +(amplified-read apparent norm — clip + msign both immune); cos 0.28 with BP-equivalent CE — +the direction-metric coffin nailed shut. +STANDING QUESTION (the honest one for the ladder): does the required dose scale with width? +x128 at C768 vs ~x1 at C512 — if dose ~ width^k the wall is soft but real; if it saturates, +this is a cure. 270M will answer. Production note: dgain is estimator-side only — model, +inference, cost, hardware story all untouched; hardware realization = per-layer nudge +amplifier gain (standard analog practice). +IN FLIGHT: fw135m_dg128 (bsign + dgain_top 128, s250000 -> 440k, 3-GPU, ~22h). Bars: +untreated 3.3839 | BP 3.0902. Success = endpoint ~3.15-3.18, gap back in the 72M band. + ### RESULT 71 (2026-07-29): DOSE CURVE — logarithmic recovery (34/50/63% at x4/8/16), DISPLACEMENT-EQUIVALENCE LAW triple-confirmed, cos retired as a metric; TREATMENT VALIDATION LAUNCHED (bsign + dg8top, full remaining segment). -- cgit v1.2.3