From 77889401c95c75a591f69e51caf7617bc848f0cb Mon Sep 17 00:00:00 2001 From: Yuren Hao Date: Tue, 28 Jul 2026 09:27:18 -0500 Subject: =?UTF-8?q?RESULT=2064:=20=E9=9D=9Eplateau=E8=80=8C=E6=98=AF?= =?UTF-8?q?=E6=81=92=E9=80=9F=E6=B3=84=E6=BC=8F(=E5=AE=BD=E5=BA=A6=C3=974-?= =?UTF-8?q?5,=E9=80=80=E7=81=AB=E8=BF=BD=E4=B8=8D=E4=B8=8A);=20=E5=85=A8?= =?UTF-8?q?=E5=AB=8C=E7=96=91=E8=87=B3=E5=88=86=E8=BE=A8=E7=8E=87=E6=B4=97?= =?UTF-8?q?=E6=B8=85(bias=E2=89=A4sem,msign=E6=97=A0=E6=94=BE=E5=A4=A7,?= =?UTF-8?q?=CF=83=E4=B8=A4=E8=BE=B9=E5=90=8C=E4=B8=80/BP=E6=9B=B4=E5=93=8D?= =?UTF-8?q?);=2072M=E8=B4=A6=E6=9C=ACn=3D3=E5=90=88=E5=8D=B7(BP=203.2882?= =?UTF-8?q?=C2=B10.006,bsign=20gap+0.076);=20=E7=A3=81=E7=9B=98=E6=BB=A1?= =?UTF-8?q?=E4=BA=8B=E6=95=85+458GB=E4=BF=AE=E5=89=AA(=E6=B8=85=E5=8D=95?= =?UTF-8?q?=E8=90=BD=E6=A1=A3);=20amp=E9=99=8D=E7=BA=A7=E6=AC=A1=E8=A6=81(?= =?UTF-8?q?~12%);=20=E6=B3=84=E6=BC=8F=E7=8E=87=E9=85=8D=E5=AF=B9=E6=8E=A2?= =?UTF-8?q?=E9=92=88=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 | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+) (limited to 'docs') diff --git a/docs/campaign/CASCADE_ABLATION_PLAN.md b/docs/campaign/CASCADE_ABLATION_PLAN.md index 1606dd2..a73f28b 100644 --- a/docs/campaign/CASCADE_ABLATION_PLAN.md +++ b/docs/campaign/CASCADE_ABLATION_PLAN.md @@ -568,6 +568,28 @@ direction), not training-under-fault; wave-2 = co-training with faults injected +### RESULT 64 (2026-07-28): NOT A PLATEAU — A CONSTANT-RATE RELATIVE LEAK, 4-5x wider at +C768; every named suspect cleared AT PROBE RESOLUTION; the leak-rate probe is the new instrument. +LEDGERS CLOSED: 72M BP n=3: 3.2884/3.2943/3.2820 -> mean 3.2882+-0.006. Honest 72M gap: +bsign(n=2) +0.076 (7.9% ppl); cent(n=1) +0.044. — fp32 A/B (both arms died at s275000 +torch.save: DISK FULL, 100%/10T shared NFS; pruned 458GB of intermediate ckpts, 652 files, +manifest runs/PRUNE_MANIFEST_260728.txt, 39 landmarks kept): fp32@3e-3 window val_mean 3.5538 +vs amp 3.5891 vs fp32@1.5e-3 3.5843 -> amp DEMOTED to minor contributor (~12% of the deficit); +R63's "amp is THE mechanism" corrected. +SUSPECTS CLEARED (with resolution caveats): systematic bias at C768 <= sem (probe_bias nb=12: +plain ~ sem everywhere; w2_b11 0.81% vs sem 0.51% = the one 1.6x flicker); Muon amplification +NONE (msign cos 0.9987-0.9994 vs raw 0.9998-1.0000); sigma/scale runaway NONE (tok_rms +IDENTICAL to 3 decimals; BP head LOUDER: sig1 744 vs EP 620 — the louder-states trajectory is +normal training, not EP pathology). +THE SHAPE (schedule-fraction matched): 72M gap hovers +0.04..0.09 and the anneal CLOSES it +(0.087@80% -> 0.047@100%); 135M gap grows near-linearly +0.10/0.16/0.20/0.28/0.29 — a +per-progress leak, 4-5x the 72M rate, that the anneal cannot outrun; "plateau" = leak rate +catching BP's descent rate. Required per-step magnitude: ~1% CORRELATED discrepancy — exactly +at/below every probe's floor. The 0.999-cos-yet-diverging paradox is the finding. +IN FLIGHT: leak-rate paired probe — SAME weights (s250000), EP vs BP, 6k steps each, direct +DCE readout; calibrates the leak instrument for recipe search (anything that cuts measured +leak-rate is a candidate; full-run validation only after). + ### RESULT 63 (2026-07-27): MECHANISM CLOSED — THE AMP CERTIFICATE BREAKS AT WIDTH AND THE C768 CORRIDOR INVERTS MID-RUN; every 135M EP failure mode falls into place. Chain of three probes (all GPU0, hours): (1) blockcos @s150000: fp32 EP gradient PERFECT -- cgit v1.2.3