From c97c991a4c338026cc77f91d1298e7d6713ca453 Mon Sep 17 00:00:00 2001 From: Yuren Hao Date: Sat, 18 Jul 2026 22:54:58 -0500 Subject: =?UTF-8?q?RESULT=2048c:=20=CE=B8-read=E7=BB=93=E6=9E=84=E4=B8=8A?= =?UTF-8?q?=E6=97=A01/=CE=B2=E9=80=9A=E9=81=93(d=E2=88=9D=CE=B2,=CE=B2?= =?UTF-8?q?=E5=9C=A8cotangent=E9=87=8C=E6=B6=88=E6=8E=89;=E4=BA=94?= =?UTF-8?q?=E7=A7=8D=E6=95=B0=E5=AD=97=E7=B3=BB=E7=BB=BC=E5=85=A8=E9=83=A8?= =?UTF-8?q?=CE=B2-=E5=B9=B3=E5=9D=A6)=E2=86=92=E6=95=B0=E5=AD=97=E5=9F=9F?= =?UTF-8?q?=E6=97=A0=E6=A3=80=E6=B5=8B=E4=B8=8B=E7=95=8C=E6=98=AF=E7=BB=93?= =?UTF-8?q?=E6=9E=84=E6=80=A7=E7=9A=84;BBP=E7=9B=B8=E5=8F=98=E7=9C=9F?= =?UTF-8?q?=E5=AE=B6=3D=E5=B7=AE=E5=88=86=E8=AF=BB=E5=87=BA=E7=A1=AC?= =?UTF-8?q?=E4=BB=B6(T64=E5=88=97=E8=AF=BB=E5=8D=B3=E6=98=AF)=E2=86=92?= =?UTF-8?q?=CE=B2*=5FHW=E8=AE=BE=E8=AE=A1=E6=96=B9=E7=A8=8B=E6=88=90?= =?UTF-8?q?=E7=AB=8B;qb6=20buyback=E6=9C=BA=E5=88=B6=E6=94=B9=E5=88=A4?= =?UTF-8?q?=E5=BE=85=E6=9F=A5?= 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 | 19 +++++++++++++++++++ 1 file changed, 19 insertions(+) (limited to 'docs') diff --git a/docs/campaign/CASCADE_ABLATION_PLAN.md b/docs/campaign/CASCADE_ABLATION_PLAN.md index 399d8d6..268ab91 100644 --- a/docs/campaign/CASCADE_ABLATION_PLAN.md +++ b/docs/campaign/CASCADE_ABLATION_PLAN.md @@ -568,6 +568,25 @@ direction), not training-under-fault; wave-2 = co-training with faults injected +### RESULT 48c (2026-07-18): WHY the digital floor can't be BBP — the THETA-READ HAS NO 1/beta +CHANNEL, structurally. The BBP transition lives exactly where readout is DIFFERENTIAL (analog). +probe_bbp3 (6-bit STOCHASTIC rounding, fresh draw per estimate; frozen = one draw for free+ +nudged [trainer-faithful], split = independent draws [decorrelated bound]): R beta-FLAT 0.3-1.4 +in BOTH modes, identical tables — even fully decorrelated operator noise produces NO 1/beta +bulk. Mechanism: our estimator reads <-d/(beta*NBT), dO/dtheta> — d is proportional to beta, the +beta CANCELS in the cotangent; quant/rounding errors enter as RELATIVE anchor/transmission +shifts O(q), never divided by beta. The (E+-E-)/2beta catastrophic-cancellation channel simply +does not exist in this code path. => Across ALL FIVE digital ensembles probed (fp32, bf16-amp, +det-quant, stoch-quant frozen, stoch-quant split): no detectability floor; wall-1-in-sim = +window geometry + bf16 displacement cliff. WHERE BBP IS REAL: hardware whose readback IS a +differential measurement (delta-I/delta-V between phases / beta) — T64's transpose/column read +IS differential -> additive ADC noise / beta = true 1/beta bulk -> beta*_HW = nu_ADC(sqrt m + +sqrt n)/sigma1(g) stands as the design equation, calibrate nu from ENOB 8.51b. +Caveats: R is an iid-frame index; 6-bit noise is STRUCTURED (correlated with signal directions: +R<1 yet overlap 0.3-0.9, not the BBP zero) — structured noise degrades gracefully, R +conservative. qb6 beta-buyback mechanism REASSIGNED to open (not 1/beta-noise; candidate: +anchor-displacement-relative channel) — measurable later, not blocking. + ### RESULT 48b (2026-07-18): GENERALIZED BBP (amp semantics, measured-spectrum, no iid assumption; user-directed): R = spike/edge > 1 EVERYWHERE down to beta=1e-4 — EP's extra noise sits 5-50x BELOW the signal spike in bf16-amp too. The digital floor is NOT a detectability -- cgit v1.2.3