close-out: harvest-0829-phase8b-final shipped (autolens_profiling#195, #193); dashboard regenerated

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LoFQxXV9W59icpEyiGL4eG

Author: 2 people

active.md

@@ -13,22 +13,6 @@
   - euclid_strong_lens_modeling_pipeline: feature/euclid-pipeline-parity
 - next-skill: /ship_workspace (stages A+B → C → D → E in flight)
 
-## harvest-0829-phase8b-final
-- issue: https://github.com/PyAutoLabs/autolens_profiling/issues/194
-- prompt: active/harvest_0829_phase8b_final_verdict.md
-- issued: 2026-08-29
-- status: library-shipped, awaiting-merge
-- worktree: ~/Code/PyAutoLabs-wt/harvest-0829-phase8b-final
-- classification: library (autolens_profiling only — results, ledger notes, wall rates, submit exit-code guard)
-- epic: jax-inference-profiling
-- repos:
-  - autolens_profiling: feature/harvest-0829-phase8b-final
-- library-pr: https://github.com/PyAutoLabs/autolens_profiling/pull/195
-- commit: 45c3abd
-- workspace-impact: none — autolens_profiling exports no API and no library source changed; the only executable edits are wall/rates.py (additive table row), one SLURM submit header, and a SLURM-scoped guard in activate.sh
-- heart: YELLOW at ship time, acknowledgement OUTSTANDING — reasons are release-chain and in other repos ("workspace validation not passing (0 failed, 1 timeout, cloud#33229145647: autolens_test scripts/multi_dataset/delaunay_mge.py)", "release validation incomplete: no rehearsal for current source"); surfaced verbatim in the PR body for the human to answer before merge
-- next-skill: /prm (merge stays human)
-
 ## multi-plane-cross-validation
 - issue: https://github.com/PyAutoLabs/PyAutoLens/issues/714
 - prompt: active/cross_validate_multi_plane_ray_tracing.md

…ve/harvest_0829_phase8b_final_verdict.md …te/2026/08/harvest-0829-phase8b-final.mdactive/harvest_0829_phase8b_final_verdict.md renamed to complete/2026/08/harvest-0829-phase8b-final.md active/harvest_0829_phase8b_final_verdict.md renamed to complete/2026/08/harvest-0829-phase8b-final.md

@@ -1,3 +1,15 @@
+## harvest-0829-phase8b-final
+- issue: https://github.com/PyAutoLabs/autolens_profiling/issues/194
+- completed: 2026-08-29
+- library-pr: autolens_profiling#195 (merged a305a49c -> main); alongside autolens_profiling#193 W6 tail submits (merged 24155feb -> main)
+- what shipped: RAL harvest of 341978/341981/341987/341988 — 18 result JSONs; Phase 8B FINAL verdict on 39/39 arms = FALSIFIED 3/4 (F1, F3, F4), `preliminary: false`, F2 not falsified, F5 diagnostic max rel 1.06e-5, 23/39 best points |e|>1; W6 n_batch tail (MGE wall optimum n_batch=2000, 4.19 ms/eval, logZ drift -0.24 nat over the scan; delaunay leg LOST to cuFFT 25 GiB OOM reported as COMPLETED); Phase 6 NUTS probe (3.42 s/draw @ 4 lanes = 38.53 ms/eval in wall/rates.py; warm arm reaches basin but rhat 3.9 / 50% divergences -> H6.1 unsupported with MAP-sourced diagonal metric; cold arm timed out at 0:45 -> submit budget 2:00). Three DECISIONS.md entries, PROGRAMME rows + new W11, phase_08 RESULTS.md + 39-arm appendix, methods/nautilus.md rows, `activate.sh` SLURM exit-code guard (85 submits).
+- validation: scorer reproduced the verdict byte-identically in the worktree; appendix re-derived from committed JSONs; lint CI green (ruff, format, build_readme --check, check_submits --check, pytest 256 passed).
+- heart-ack: shipped + merged under human-authorised YELLOW ("1 yes", 2026-08-29) — reasons verbatim: "workspace validation not passing (0 failed, 1 timeout, cloud#33229145647: autolens_test scripts/multi_dataset/delaunay_mge.py)" and "release validation incomplete: no rehearsal for current source" — release-chain, other repos; this PR touches no library source.
+- caveats: PyAutoFit stack SHA is unverifiable from artifacts (JSONs carry only `version: 2026.8.17.1`) — "library SHAs in the results schema" filed in W11; sacct COMPLETED is not evidence (341988) — guard added.
+- human calls left open: delaunay n_batch tail redesign DECLINED 2026-08-29 ("doesn't sound worth it"); dense-metric NUTS probe from a Nautilus warm start APPROVED in principle, to be planned as a new task; slam_source_pix_nn refs 5,6 still HELD.
+
+## Original prompt
+
 # Harvest 2026-08-29: Phase 8B FINAL verdict, W6 n_batch tail, Phase 6 NUTS probe
 
 Type: feature

complete/index.md

@@ -6,7 +6,7 @@ Token-light navigation over the finished-work records (schema:
 only then grep a dated bucket. Curators: edit the band between the CURATED
 markers; everything below GENERATED is rebuilt.
 
-1210 records across 7 buckets.
+1211 records across 7 buckets.
 
 <!-- CURATED:START -->
 ## Highlights
@@ -104,6 +104,7 @@ _(curate hard-won records here — survives regeneration.)_
 - [hands-hygiene-leftovers](2026/08/hands-hygiene-leftovers.md)
 - [hands-raw-string-docstring-prefix](2026/08/hands-raw-string-docstring-prefix.md) — closed on merge
 - [harvest-0827-gate-b-pt2](2026/08/harvest-0827-gate-b-pt2.md) — 2026-08-27 RAL A100 harvest + audit (https://claude.ai/code/artifact/d9f4b0f3-52a1-4830-a9ad-11a225b77507). Ga…
+- [harvest-0829-phase8b-final](2026/08/harvest-0829-phase8b-final.md)
 - [hazard-profiling-likelihood-tier](2026/08/hazard-profiling-likelihood-tier.md) — Tier-2 likelihood profiling landed with five persistent findings and corrected NNLS continuity semantics.
 - [health-conductor-stale-verdict](2026/08/health-conductor-stale-verdict.md)
 - [heart-green-validation-ingest](2026/08/heart-green-validation-ingest.md)