record: inference-programme-ledger shipped (autolens_profiling#136+#137)

active.md

@@ -1,18 +1,5 @@
 # Active Tasks
 
-## inference-programme-ledger
-- issue: https://github.com/PyAutoLabs/autolens_profiling/issues/134
-- status: pr-open (https://github.com/PyAutoLabs/autolens_profiling/pull/135)
-- worktree: ~/Code/PyAutoLabs-wt/inference-programme-ledger
-- repos:
-  - autolens_profiling: feature/inference-programme-ledger
-- prompt: active/inference_programme_ledger.md
-- note: docs-only (results/notes/inference/ PROGRAMME.md + DECISIONS.md + LITERATURE.md).
-  The mge-lane-death claim it was recorded as disjoint from is gone — archived
-  2026-08-18 as superseded (`complete/2026/08/mge-lane-death.md`); its
-  `research/mge-lane-death` branch was never created.
-  This is the canonical copy of the 2026-08-17 human-approved inference programme.
-
 ## stored-sample-reconstruction-guard
 - issue: https://github.com/PyAutoLabs/PyAutoFit/issues/1486
 - status: library-dev — WORKSPACE HALF SHIPPED; the PyAutoFit hardening (#1486) is what remains

complete/2026/08/inference-programme-ledger.md

@@ -0,0 +1,69 @@
+Committed the human-approved 2026-08-17 inference-methods programme into its
+canonical home, `autolens_profiling/results/notes/inference/`, per the
+programme's own knowledge rules (no knowledge lives only in chat histories or
+artifacts). Documentation only — no runs, no scripts, no source changes.
+
+- **PRs:** autolens_profiling#136 (merged `3be8158` — PROGRAMME.md +
+  DECISIONS.md + LITERATURE.md, adapted from the approved artifact, with a
+  maintained phase/gate state table and CP-1 marked complete) and
+  autolens_profiling#137 (merged `5ce0243` — folded in the laptop session's
+  richer content; closed issue #134).
+- **Duplicate-work trap:** the task was started twice. A laptop session had
+  already opened autolens_profiling#135 (`feature/inference-programme-ledger`,
+  registered in active.md) when a mobile/cloud session redid the task on
+  `claude/inference-methods-autolens-w86nz9` and merged #136 without checking
+  the registry first. The versions were complementary — #136 had the fuller
+  PROGRAMME body (995 vs 558 lines), #135 the richer CP-1 evidence
+  (test-pin values, 538-test suite, fence-slope-halving science impact,
+  truth-bars-unaffected note), the deeper LITERATURE set (~25 refs incl. the
+  optimizer-termination section), and the Future follow-ups section. #137
+  reconciled them; #135 closed unmerged, nothing lost. Lesson: check
+  active.md's PR/status line before re-issuing an in-flight task from another
+  device.
+- **Leftovers:** laptop worktree `~/Code/PyAutoLabs-wt/inference-programme-ledger`
+  and remote branch `feature/inference-programme-ledger` are stale — next
+  repo_cleanup pass.
+- **What the ledger seeds:** DECISIONS.md carries the plan approval (verbatim
+  review notes) + CP-1 completion (PositionsLH accumulation fix,
+  PyAutoLens#699/PR#700). Next programme steps per PROGRAMME.md's critical
+  path: CP-2 blackjax ≥1.6.2 upgrade + mainline NSS smoke (laptop GPU), then
+  CP-3 Prodigy reliability scan; Phase 0 (b)(c)(e) outstanding.
+
+## Original prompt
+
+# Commit the inference-methods programme plan + knowledge ledger into autolens_profiling
+
+Type: docs
+Target: autolens_profiling
+Repos:
+- @autolens_profiling
+Difficulty: small
+Autonomy: supervised
+Priority: high
+Status: draft
+
+The 2026-08-17 inference-methods planning pass (human-approved same day, with
+notes) produced a full phased R&D programme: state of play, phases 0-13 with
+decision gates A-F, benchmark/result schema v2, method-card knowledge structure,
+source-change list, risk register, critical path, and a hypothetical decision
+tree. Human review notes: (1) PositionsLH bug — verified, fix shipped as
+PyAutoLens#699/PR#700; (2) BlackJAX NS "was fastest or comparable on MGE and may
+scale better — worth rerunning" (Phase 2 stands); (3-6) approved as planned.
+
+Per the programme's own knowledge rules, the canonical copy must live in
+autolens_profiling, not in chat history or an artifact.
+
+Task: create `results/notes/inference/` containing:
+
+- `PROGRAMME.md` — the full plan (markdown adaptation of the approved artifact),
+  maintained as phases execute.
+- `DECISIONS.md` — append-only gate log, seeded with the 2026-08-17 plan
+  approval + review notes and the CP-1 completion (PositionsLH fix PR).
+- `LITERATURE.md` — the external references gathered during planning (blackjax
+  1.6 nested sampling merge, NSS paper, GIGA-Lens/Herculens, NUTS diagnostics,
+  informed-start SMC, MCLMC/MAMS, Pathfinder, proximal smoothing), one lesson
+  line each.
+
+No profiling runs, no scripts, no source changes — documentation only. Keep
+`ruff check .` / `ruff format --check .` green (no .py files added) and do not
+touch the auto-generated README tables.

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.
 
-1007 records across 7 buckets.
+1008 records across 7 buckets.
 
 <!-- CURATED:START -->
 ## Highlights
@@ -68,6 +68,7 @@ _(curate hard-won records here — survives regeneration.)_
 - [hygiene-detail-flag](2026/08/hygiene-detail-flag.md)
 - [hygiene-gone-ref-counter](2026/08/hygiene-gone-ref-counter.md) — the prompt was never issued — the defects were fixed under other issues, see below
 - [hygiene-howto-refs-docstrings](2026/08/hygiene-howto-refs-docstrings.md)
+- [inference-programme-ledger](2026/08/inference-programme-ledger.md)
 - [interferometer-delaunay-flaky-fitexception](2026/08/interferometer-delaunay-flaky-fitexception.md)
 - [interferometer-start-here-integrate-oom](2026/08/interferometer-start-here-integrate-oom.md)
 - [intra-family-dep-floors](2026/08/intra-family-dep-floors.md)