From 574e63cd4cf988284cd029bc0e0b708ec23d4b62 Mon Sep 17 00:00:00 2001 From: Yuren Hao Date: Wed, 29 Jul 2026 06:47:48 -0500 Subject: =?UTF-8?q?RESULT=2070:=20=E9=A6=96=E6=89=B9=E7=9C=9F=E6=9D=A0?= =?UTF-8?q?=E6=9D=86=20=E2=80=94=20dg4top+34%(cos0.83=E4=B8=8BCE=E8=83=9C?= =?UTF-8?q?=E5=87=BA),=CE=B2=E6=94=B6=E7=9B=8A=3D100%=E4=BD=8D=E7=A7=BB0%?= =?UTF-8?q?=E5=8A=9B(dg2all=E2=89=A1b6d=E9=80=90=E4=BD=8D),ffn~=E4=B8=80?= =?UTF-8?q?=E5=8D=8A+=E4=BA=A4=E4=BA=92=E9=A1=B9;=20=E6=9C=BA=E5=88=B6?= =?UTF-8?q?=E7=94=BB=E5=83=8F=3D=E9=A1=B6=E5=8D=8Aattn+ffn=E7=9A=84?= =?UTF-8?q?=E4=BD=8D=E7=A7=BB=E9=98=88=E5=80=BC=E7=8E=B0=E8=B1=A1,?= =?UTF-8?q?=E7=BA=AF=E4=BC=B0=E8=AE=A1=E5=99=A8=E4=BE=A7=E6=B2=BB=E6=B3=95?= =?UTF-8?q?=E5=AD=98=E5=9C=A8;=20=E5=89=82=E9=87=8F=E6=9B=B2=E7=BA=BFdg8/d?= =?UTF-8?q?g16+=E5=8F=A0=E5=8A=A0=E8=87=82=E5=9C=A8=E9=A3=9E?= 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 | 18 ++++++++++++++++++ 1 file changed, 18 insertions(+) diff --git a/docs/campaign/CASCADE_ABLATION_PLAN.md b/docs/campaign/CASCADE_ABLATION_PLAN.md index 0df3e72..efe5856 100644 --- a/docs/campaign/CASCADE_ABLATION_PLAN.md +++ b/docs/campaign/CASCADE_ABLATION_PLAN.md @@ -568,6 +568,24 @@ direction), not training-under-fault; wave-2 = co-training with faults injected +### RESULT 70 (2026-07-29): FIRST REAL LEVERS — displacement amplification recovers the +leak (dg4top +34% at cos-0.83 price, CE judges); the beta benefit factorizes to 100% +DISPLACEMENT, 0% force; ffn carries ~half with an attn-ffn interaction term. +Battery 6: leak_dg4top closes +34% (top-half state-formation d x4; despite gate cos 0.83 — +FA-logic vindicated, CE is the judge); leak_dg2all +24% with tail IDENTICAL to b6d (3.5558 +== 3.5558): doubling displacement-only reproduces doubling beta exactly -> the 1/beta dose +response is entirely the DISPLACEMENT channel (force/SNR contributes nothing measurable); +leak_mixffn +48% (attn 67 + ffn 48 > 100: overlapping/non-additive at C768, unlike R44's +clean 72M additivity — an interaction term exists). +MECHANISM PORTRAIT (refined): a displacement-threshold phenomenon distributed across +top-half attn AND ffn nonlinearities — response components that finite probe displacement +under-reaches, linearly redeemed by amplifying displacement through ANY route. Not softmax +logits (R69). Pure estimator-side treatment exists (dgain: model function, inference, cost +all untouched; hardware = per-layer nudge amplifier gain, standard analog practice). +Battery 7 IN FLIGHT: dose curve dg8top/dg16top + stack dg4top-x-beta6e-3. If the curve +reaches 60-80% recovery before contamination turns, the 135M treatment recipe = bsign + +dgain-top(opt) at 1.0x, full-run validation next. + ### RESULT 69 (2026-07-29): SATURATION-BLINDNESS REFUTED BY ITS OWN TREATMENT TEST; the leak survives exclusion #11; dgain (displacement-only amplification) is the live arm. Battery 4 (beta dose): leak INVERSE in beta — 1.5e-3: 0.0615, 3e-3: 0.0462, 6e-3: 0.0352 -- cgit v1.2.3