Jammy2211 · Oct 8, 2026 · Oct 8, 2026
diff --git a/campaigns.yaml b/campaigns.yaml
@@ -77,10 +77,10 @@ campaigns:
   title: Measurement reliability and profiling tools
   status: active
   updated: '2026-10-08'
-  next: 'timing_noise_audit fix phases (1) and (2) done (autolens_profiling#404 merged; #405 open, awaiting /prm). Correction
-    2026-10-08: the live Pulse lens feed is dashboard/catalogue.json (profiling-summary@2, registry switched 2026-10-05
-    in 3c7bed2), whose producer hard-codes qualified: false, so the v1 qualify() gap never reached Pulse live and fix (2)
-    needed no Pulse change. Next: fix phase (3) INCONCLUSIVE for go/lever rules. Other tasks unchanged and unissued.'
+  next: 'timing_noise_audit fix phases (1)-(4) MERGED 2026-10-08 (autolens_profiling#404, #405, #406, #407). The live Pulse
+    lens feed is the v2 catalogue.json (qualified hard-false), so none needed a Pulse change. Next: fix phase (5) median
+    headline + GPU-only marker, then (6) warm-up flag + witness band; leftovers phase 3b (C1/C3/C4/C5), Holm/Bonferroni
+    policy, gpu_bottleneck_map round bootstrap. Other tasks unchanged and unissued.'
   evidence: https://github.com/PyAutoLabs/autolens_profiling/blob/main/wiki/campaigns/measurement_tools.md
 - id: autofit
   title: PyAutoFit profiling bootstrap
@@ -139,14 +139,14 @@ tasks:
   priority: high
   path: tasks/timing_noise_audit.md
   updated: '2026-10-08'
-  next: 'Fix phase (2) PR OPEN 2026-10-08 as autolens_profiling#405 (awaiting /prm; Heart YELLOW reasons human-acknowledged):
-    dashboard qualify() leaves laptop rows, rows without a load average and hpc rows without a host unqualified with a
-    reason; a within-2x comparison is flat only when both endpoints carry a repeat summary (none do yet), else insufficient;
-    drifted/improved keep status with a single-sample caveat (human decision 2026-10-08). Effect on summary.json: qualified
-    2 -> 2, flat 4 -> 0, insufficient 135 -> 139. Not a Pulse contract change (live feed is v2 catalogue.json, qualified
-    hard-false). Next after merge: fix phase (3) INCONCLUSIVE for go/lever rules, tie sets instead of argmax.'
+  next: 'Fix phases (2)-(4) MERGED 2026-10-08 under human /prm: #405 dashboard qualification + drift wording; #406 A/B go/lever
+    rules gain INCONCLUSIVE + tie sets (shared ab_verdict.py; no published go/no-go/NO_LEVER changed; 7 of 9 committed C10
+    sweeps are tie sets); #407 paired whole-round bootstrap (shared round_bootstrap.py; RAL CPU/A100 calls unchanged; one
+    laptop C6 row GO -> INCONCLUSIVE). Issue #362 stays open. Next: fix phase (5) P8/P9 (median headline, GPU-only marker),
+    then (6) P3/P5 (authorized --auto; not started at session end); leftovers phase 3b (C1/C3/C4/C5 lack intervals),
+    multiple-comparison policy, gpu_bottleneck_map.'
   issue: https://github.com/PyAutoLabs/autolens_profiling/issues/362
-  evidence: https://github.com/PyAutoLabs/autolens_profiling/pull/405
+  evidence: https://github.com/PyAutoLabs/autolens_profiling/pull/407
 - id: mge_likelihood_breakdown_steps_are_cumulative_an
   campaign: measurement-tools
   title: MGE likelihood_breakdown steps are cumulative and `linear_gaussians` is reported as 0

diff --git a/dashboard.html b/dashboard.html
diff --git a/dashboard.md b/dashboard.md
@@ -1,6 +1,6 @@
 # PyAutoPulse — profiling dashboard
 
-<!-- pulse-campaigns:652e6582f3ead66f68ae1120a79f68f1c18219dbc4ffee221291672133f5fc79 -->
+<!-- pulse-campaigns:885c41acf471ee25bdaad4bb819d702e5d3e8e19f7b3662b9e9f46a04b538c6d -->
 ## Profiling Check In
 
 <details><summary>Full check-in prompt</summary>
@@ -133,13 +133,13 @@ Reviewed 2026-10-04. No new results since the pin. Over-sampling phase 1: add --
 
 <details><summary>Measurement reliability and profiling tools — active</summary>
 
-Reviewed 2026-10-08. timing_noise_audit fix phases (1) and (2) done (autolens_profiling#404 merged; #405 open, awaiting /prm). Correction 2026-10-08: the live Pulse lens feed is dashboard/catalogue.json (profiling-summary@2, registry switched 2026-10-05 in 3c7bed2), whose producer hard-codes qualified: false, so the v1 qualify() gap never reached Pulse live and fix (2) needed no Pulse change. Next: fix phase (3) INCONCLUSIVE for go/lever rules. Other tasks unchanged and unissued.
+Reviewed 2026-10-08. timing_noise_audit fix phases (1)-(4) MERGED 2026-10-08 (autolens_profiling#404, #405, #406, #407). The live Pulse lens feed is the v2 catalogue.json (qualified hard-false), so none needed a Pulse change. Next: fix phase (5) median headline + GPU-only marker, then (6) warm-up flag + witness band; leftovers phase 3b (C1/C3/C4/C5), Holm/Bonferroni policy, gpu_bottleneck_map round bootstrap. Other tasks unchanged and unissued.
 
 [Campaign evidence](https://github.com/PyAutoLabs/autolens_profiling/blob/main/wiki/campaigns/measurement_tools.md)
 
 ### Active tasks
 
-- [Audit timing tests and profiling gates for measurement noise](https://github.com/PyAutoLabs/PyAutoPulse/blob/main/tasks/timing_noise_audit.md) — active: Fix phase (2) PR OPEN 2026-10-08 as autolens_profiling#405 (awaiting /prm; Heart YELLOW reasons human-acknowledged): dashboard qualify() leaves laptop rows, rows without a load average and hpc rows without a host unqualified with a reason; a within-2x comparison is flat only when both endpoints carry a repeat summary (none do yet), else insufficient; drifted/improved keep status with a single-sample caveat (human decision 2026-10-08). Effect on summary.json: qualified 2 -&gt; 2, flat 4 -&gt; 0, insufficient 135 -&gt; 139. Not a Pulse contract change (live feed is v2 catalogue.json, qualified hard-false). Next after merge: fix phase (3) INCONCLUSIVE for go/lever rules, tie sets instead of argmax.
+- [Audit timing tests and profiling gates for measurement noise](https://github.com/PyAutoLabs/PyAutoPulse/blob/main/tasks/timing_noise_audit.md) — active: Fix phases (2)-(4) MERGED 2026-10-08 under human /prm: #405 dashboard qualification + drift wording; #406 A/B go/lever rules gain INCONCLUSIVE + tie sets (shared ab_verdict.py; no published go/no-go/NO_LEVER changed; 7 of 9 committed C10 sweeps are tie sets); #407 paired whole-round bootstrap (shared round_bootstrap.py; RAL CPU/A100 calls unchanged; one laptop C6 row GO -&gt; INCONCLUSIVE). Issue #362 stays open. Next: fix phase (5) P8/P9 (median headline, GPU-only marker), then (6) P3/P5 (authorized --auto; not started at session end); leftovers phase 3b (C1/C3/C4/C5 lack intervals), multiple-comparison policy, gpu_bottleneck_map.
 - [MGE likelihood_breakdown steps are cumulative and `linear_gaussians` is reported as 0](https://github.com/PyAutoLabs/PyAutoPulse/blob/main/tasks/mge_likelihood_breakdown_steps_are_cumulative_an.md) — ready: Unchanged and unissued (no GitHub refs, 2026-10-04). Select one bounded step when measurement-tools is prioritised.
 - [A gradient-cost probe: forward vs `value_and_grad` ms/eval and a strict FD check, on any registry cell](https://github.com/PyAutoLabs/PyAutoPulse/blob/main/tasks/gradient_cost_probe.md) — ready: Unissued. New context 2026-10-07: Heart unit-test timing flagged PyAutoLens forward-gradient tests +150%/+96% after PyAutoLens#768 rewrote that test file; not a controlled comparison, and this probe is the right instrument. Select one bounded step when prioritised.
 - [Numba breakdown harness: perturb the instance so the operated-matrix memo cannot hide a step](https://github.com/PyAutoLabs/PyAutoPulse/blob/main/tasks/numba_breakdown_harness_memo_blind.md) — ready: Unchanged and unissued (PyAutoArray#496 closed 2026-08-27). Select one bounded step when prioritised.

diff --git a/tasks/timing_noise_audit.md b/tasks/timing_noise_audit.md
@@ -54,6 +54,16 @@ The original task below is retained verbatim. Current scheduling state lives in
 - Carry-forward: when the v2 catalogue starts qualifying records (setup-baseline reference-host decision), it must reuse the producer's `is_reference_host_class` rule rather than a separate one.
 - Next: /prm the PR (judge tier, human merge); then fix phase (3) — INCONCLUSIVE for go / lever rules, tie sets instead of argmax.
 
+
+### Phases 3–5 merged 2026-10-08 (fix phases 2–4)
+
+- [autolens_profiling#405](https://github.com/PyAutoLabs/autolens_profiling/pull/405) merged 19:46Z (f6e6982); Mind record `complete/2026/10/timing-noise-audit-p3-qualify-drift.md` (d774cbb1).
+- [autolens_profiling#406](https://github.com/PyAutoLabs/autolens_profiling/pull/406) merged 20:19Z (94861700): shared `ab_verdict.py` (`ab_rule_verdict`, `tie_set`, `paired_block_ratio_interval`) wired into C6, C7, P2, C10. Re-judged committed rows: no go / no-go / NO_LEVER call changed; 7 of 9 committed C10 "best" picks are tie sets (IP-4a's 2.37x leader ties with four others; the human decision not to ship it stands). Flagged decision: C10 vmap rows measured for the tie set's `point_leader` only. Mind record f55f220e.
+- [autolens_profiling#407](https://github.com/PyAutoLabs/autolens_profiling/pull/407) merged 20:43Z (8be806cc): shared `round_bootstrap.py` (paired whole-round resampling, `effective_n = n_rounds`) in seven cells. Witness: iid CI 0.14–0.39x as wide as the round CI; 40-seed coverage 0.85 vs 0.33 at nominal 0.90. RAL CPU / A100 calls unchanged; laptop C6 solved `rev_analytic` GO -> INCONCLUSIVE ([0.805, 0.854] vs 0.85). The review returned FINDINGS (label strings); the fix commit 63642d52 was re-reviewed by the main session before merge.
+- Heart was YELLOW throughout with an organism-wide manifest-drift + no-rehearsal set; the human acknowledged that set for these PRs on 2026-10-08.
+- #407 close-out: Mind record `complete/2026/10/timing-noise-audit-p5-round-bootstrap.md` (a9eb5864). Fix phases (5) and (6) NOT started at session end (no branch, Mind entry or plan comment); resume at start_dev with slug `timing-noise-audit-p6-median-headline-gpu-marker` from main 8be806c.
+- Leftovers recorded in the audit note: phase 3b (C1/C3/C4/C5 have no interval), multiple-comparison policy (Holm/Bonferroni), `gpu_bottleneck_map` still per-call. Next: fix phase (5) P8/P9, then (6) P3/P5 (human authorized both under --auto 2026-10-08).
+
 ---
 
 # Audit timing tests and profiling gates for measurement noise