From 4cf3811692022eb6f7ba89d5445661d078afce32 Mon Sep 17 00:00:00 2001 From: Yuren Hao Date: Wed, 22 Jul 2026 01:41:26 -0500 Subject: =?UTF-8?q?RESULT=2058:=20bsign=20s2=E5=AE=8C=E8=B5=9B3.3924=20?= =?UTF-8?q?=E2=80=94=20=E7=A8=B3=E5=AE=9A=E6=80=A7n=3D2=E5=B0=81=E5=8D=B0(?= =?UTF-8?q?=E4=B8=A4=E8=B7=91=E5=85=A8=E7=A8=8B=E9=9B=B6=E4=BA=8B=E4=BB=B6?= =?UTF-8?q?),CE=E5=9D=87=E5=80=BC3.3637=C2=B10.029(s1=E5=8F=88=E6=98=AF?= =?UTF-8?q?=E5=B9=B8=E8=BF=90=E6=8A=BD=E6=A0=B7,dip=E7=AC=AC=E5=9B=9B?= =?UTF-8?q?=E5=92=AC);=2072M=E5=85=A8=E8=A1=A8n=3D1=E6=95=B0=E5=AD=97?= =?UTF-8?q?=E5=B8=A6=C2=B10.03-0.06=E7=A7=8D=E5=AD=90=E5=B8=A6,bsign-vs-BP?= =?UTF-8?q?=20gap=E5=BE=85=E5=8D=8F=E4=BD=9C=E8=80=85BP=20seeds;=20135M?= =?UTF-8?q?=E8=BF=813=E5=8D=A1DDP?= 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 | 10 ++++++++++ 1 file changed, 10 insertions(+) (limited to 'docs/campaign') diff --git a/docs/campaign/CASCADE_ABLATION_PLAN.md b/docs/campaign/CASCADE_ABLATION_PLAN.md index d36ed9c..55b6977 100644 --- a/docs/campaign/CASCADE_ABLATION_PLAN.md +++ b/docs/campaign/CASCADE_ABLATION_PLAN.md @@ -568,6 +568,16 @@ direction), not training-under-fault; wave-2 = co-training with faults injected +### RESULT 58 (2026-07-22): BSIGN SEED-2 — STABILITY REPLICATED (n=2, zero events both +runs); CE mean 3.3637 +- 0.029, s1 was the lucky draw (dip-statistics bite #4). +fw72m_bsign_s2 DONE 3.3924: full 234k, d0/g0/r0 end to end under the res-alone gate — the +corridor-holds claim is now two-for-two; the load-bearing stability result is sealed. +CE honest ledger at 72M now: bsign mean 3.3637 (n=2) | cent 3.3318 (n=1) | plain2 3.3437 +(n=1, stalled) | BP 3.2884 (n=1). Single-seed spread at 72M is +-0.03-0.06 -> every n=1 +number above carries that band; the bsign-vs-BP gap is UNQUOTABLE until BP seeds land +(collaborator first task, BASELINE_SPEC). Flagship pair (cent s1 vs BP s1, DCE 0.043) stands +as a sealed-run fact. 135M migrated to 3-GPU DDP (B8x3) from s60000 after s2 freed the pair. + ### RESULT 57 (2026-07-21): RATCHET VERDICT — pure retreat cannot outrun the bias-driven sink; the staircase IS the plain-lineage ceiling trajectory, and it chases beta to the bf16 floor. The trio closes: debiasing is the only 1.0x recipe that holds. -- cgit v1.2.3