From d34fa415c3151e1023b66eb17b6c0237d6e75cf7 Mon Sep 17 00:00:00 2001 From: Claude Date: Tue, 17 Nov 2026 21:57:23 +0000 Subject: [PATCH] docs: the slowdown at 046ce0be is not a regression (Captain Bob) The direct loop before 046ce0be was architecturally incorrect. The fabric wiring removed it and exposed the slowdowns, and the instruments (ASan, UBSan, perf) locate them. FABRIC-3 XLVIII.24 and .25 and both 11-17 plans are corrected in place, and every measurement is kept. Mechanism A is the first speed-up found, not a fix. Co-Authored-By: Claude Opus 5.5 Claude-Session: https://claude.ai/code/session_017RzZgrf7V7Zk2iczBXDRbm --- docs/fabric/FABRIC-3.md | 17 +++++++++++------ .../plans/2026-11-17-the-fabric-step-cost.md | 6 ++++++ .../2026-11-17-the-regression-at-046ce0be.md | 6 ++++++ 3 files changed, 23 insertions(+), 6 deletions(-) diff --git a/docs/fabric/FABRIC-3.md b/docs/fabric/FABRIC-3.md index a80a3391..d59396cb 100644 --- a/docs/fabric/FABRIC-3.md +++ b/docs/fabric/FABRIC-3.md @@ -6668,17 +6668,22 @@ and memorize frequently so we can pick up where we left off"). `word_count=377` (the two telling words); `dict_hash` changed once, for the new cell. The boot took 142 s against Phase 1's 153 s: no measurable cost. **Corrected 2026-11-17:** "no measurable cost" was measured against Phase 1, which was already 1.5 to - 2.4 times slower than f10d2bc8. The regression came in at 046ce0be: every fabric step walked every - place. See 25. + 2.4 times slower than f10d2bc8. That slowdown was exposed at 046ce0be, when the fabric wiring + replaced the direct loop, which was architecturally incorrect: every fabric step walks every place. + See 25. - **Next:** Phase 7, the loops on these records (decay, the rolling window, pipelining, inference, L8). -### 25. The regression found; the fault dump; the one adaptive heartbeat (2026-11-17) +### 25. The slowdown located; the fault dump; the one adaptive heartbeat (2026-11-17) -**The regression.** aarch64 had run out of its 570 s in POST since 37469f6c (990 cases). Captain Bob: +**The slowdown.** Captain Bob: "It was architecturally incorrect. That's why we went back and did this +wiring; to eliminate the defective architecture and locate slowdowns" -- with ASan, UBSan, perf et cetera. +The direct loop before 046ce0be was the defective architecture. The fabric wiring removed it, and the +slowdown is what that exposed: not a regression. aarch64 had run out of its 570 s in POST since 37469f6c +(990 cases). Captain Bob: "find out when that happened, and actually start all over no matter how far back you need to go"; "Find the TCG cost first". -- **Found at 046ce0be**, the mesh boot to spec. The protocol logs give the stage times, for the same +- **Exposed at 046ce0be**, the mesh boot to spec. The protocol logs give the stage times, for the same 539-case POST: | Arch | f10d2bc8 | 046ce0be | @@ -6700,7 +6705,7 @@ the TCG cost first". The cost method of record (five alternating runs, as for HEAT-OP and A2) was not used after 046ce0be. - **Write-up:** `docs/v4.0.0/plans/2026-11-17-the-regression-at-046ce0be.md`. **Design:** `2026-11-17-the-fabric-step-cost.md`. Mechanism A, each pass walking only the places that hold a node, - is ruled and follows the heartbeat. + is ruled and follows the heartbeat. It is the first speed-up the instruments found, not a fix. **The CI directive** (Captain Bob, the same day): - POST is as v3's; diff --git a/docs/v4.0.0/plans/2026-11-17-the-fabric-step-cost.md b/docs/v4.0.0/plans/2026-11-17-the-fabric-step-cost.md index ec756845..d634eb46 100644 --- a/docs/v4.0.0/plans/2026-11-17-the-fabric-step-cost.md +++ b/docs/v4.0.0/plans/2026-11-17-the-fabric-step-cost.md @@ -1,5 +1,11 @@ # The fabric step's cost -- the regression at 046ce0be fixed -- design, for ruling +**Corrected 2026-11-17, Captain Bob:** this is not a regression. "It was architecturally incorrect. That's +why we went back and did this wiring; to eliminate the defective architecture and locate slowdowns" -- with +ASan, UBSan, perf et cetera. The direct loop before 046ce0be was the defective architecture. The fabric +wiring removed it and exposed the slowdowns, and the instruments locate them. The measurements below stand; +where this document says "regression", read "the slowdown the correct architecture exposed". + Written 2026-11-17. Captain Bob, asked how to handle aarch64 running out of its 570 s in POST: **"Fix the regression"**. This is treated as a correctness regression, not as tuning. The background is in `2026-11-17-the-regression-at-046ce0be.md`. diff --git a/docs/v4.0.0/plans/2026-11-17-the-regression-at-046ce0be.md b/docs/v4.0.0/plans/2026-11-17-the-regression-at-046ce0be.md index c22c7b52..cd73962c 100644 --- a/docs/v4.0.0/plans/2026-11-17-the-regression-at-046ce0be.md +++ b/docs/v4.0.0/plans/2026-11-17-the-regression-at-046ce0be.md @@ -1,5 +1,11 @@ # The regression at 046ce0be -- write-up, for ruling +**Corrected 2026-11-17, Captain Bob:** this is not a regression. "It was architecturally incorrect. That's +why we went back and did this wiring; to eliminate the defective architecture and locate slowdowns" -- with +ASan, UBSan, perf et cetera. The direct loop before 046ce0be was the defective architecture. The fabric +wiring removed it and exposed the slowdowns, and the instruments locate them. The measurements below stand; +where this document says "regression", read "the slowdown the correct architecture exposed". + Written 2026-11-17, after reading the record again in full (notes: `.claude/memory/reading-2026-11-17.md`). **Nothing here is built, reverted or reset.** Every decision below waits for Captain Bob.