Work Classification

Library (organ source + docs; no workspace scripts, no notebooks).

Affected Repositories

• PyAutoMind (primary)
• PyAutoBrain
• PyAutoCortex

Branch Survey

Suggested branch: feature/mind-post-cortex-p2-science-residue
Worktree root: ~/Code/PyAutoLabs-wt/mind-post-cortex-p2-science-residue/

Worktree claim note: worktree_check_conflict reports PyAutoMind claimed by mind-post-cortex-p1. That claim is a ledger-only branch (worktree: -) whose commits are already on main (49d0b530); the task is merely not closed out yet. Not a real conflict — phase 2 proceeds.

Implementation Steps

1. epics.md — cortex-half: keys. Delete the key from jax-inference-profiling, cluster-strong-lensing, graphical-ep, euclid-dr1-prep. For the three whose Mind half is live, add science half: PyAutoCortex epics.md#<slug> to notes:. Update the schema paragraph (line 11) if it names the key. Nothing reads it: _intake.py _EPIC_FIELDS = ("title","ledger","status","notes").
2. epics.md — retire jax-inference-profiling. Detach draft/feature/workspaces/phase_5_dev_leg_prepare_the_mesh.md (drop Epic:/Phase:). Set the entry's status: to COMPLETE — moved to Cortex 2026-09-01 and run lifecycle.py epics --retire; the ledger is external (autolens_profiling/…/PROGRAMME.md) so only the entry text is archived.
3. Lane:. Delete the ### Lane: — where the work can run section from REFERENCE.md, the lane: line from queue.md's schema and its paragraph, and the Lane: local-dev header from draft/maintenance/autolens_profiling/legacy_point_output_sweep.md (that file moves to the Cortex under step 6 anyway). Cortex REFERENCE.md:81 already documents Lane: — no Cortex edit needed. Brain's _batch.py lane filter is left alone.
4. queue.md kinds. Schema becomes kind: prompt | retired; drop the epic-slice / theme-sweep bullets and the "Three kinds" framing; drop lane:. Collapse entries 2, 3, 4, 6, 8, 9 (kind: retired) into a short "Retired" footer. Mirror the REFERENCE.md "queue and the batch records" paragraph.
5. experiment/ work-type. Mind: ROUTING.md table row, REFERENCE.md taxonomy table + tree + good-paths example, AGENTS.md work-type list, scripts/status.sh WORK_TYPES, scripts/spawn.py MIND_WORK_TYPES, docs/pyautobrain/spawn_spec.md rule 6. Brain: AUTONOMY.md cap table row first (doctrine edit), then agents/faculties/sizing/_sizing.py WORK_TYPES + WORK_TYPE_AUTONOMY_CAPS, agents/conductors/intake/_intake.py TYPE_SIGNALS + TYPE_PRECEDENCE, agents/conductors/intake/INTAKE_TAXONOMY.md, agents/conductors/intake/AGENTS.md, agents/conductors/feature/_feature.py (two ("research", "experiment") tuples), agents/conductors/feature/MIND_TAXONOMY.md, agents/conductors/feature/AGENTS.md. No Brain test fixture enumerates it (checked). Leave the research cap and the samplers faculty's unrelated "experiment tier" vocabulary alone.
6. Run-shaped prompts → Cortex phases. slope_hierarchy_methods_writeup.md → phases/slope_hierarchy/methods_writeup.md (Phase 2); cluster_gradient_search_benchmark.md → phases/inference_programme/cluster_gradient_search_benchmark.md (Phase 17); multiband_compile_census_completion.md → phases/inference_programme/multiband_compile_census_completion.md (Phase 18); legacy_point_output_sweep.md → phases/inference_programme/legacy_point_output_sweep.md (Phase 19). Each keeps its prose as the phase body, gains State: planned + Migrated-from:, and is git rm'd from Mind. ep_campaign.md stays in Mind — see "Deviation" below.
7. 2026-08-31-am trio. batches/packets/2026-08-31-am.html is byte-identical to the Cortex copy; batches/reviews/2026-08-31-am.md is a strict subset of it. The record itself carries three facts the Cortex transcription dropped (the dev member euclid-phase2-pr46, the pre-framework usage-window/heart-ack/expected-effects n/a lines, and the Mind commit refs for the packet fix) — port those into the Cortex record's notes:, correct its "the Mind original stays intact" sentence, then git rm all three Mind files.

Deviation from the prompt (item 6, one of five)

draft/research/graphical_ep/ep_campaign.md is not moved. It is not run-shaped: it is the graphical-ep campaign phase map, and of its six phases 1, 2, 5 and 6 are Mind development work — five Mind drafts still declare Epic: graphical-ep (graphical_scoping.md, ep_scoping.md, ep_lbfgs_jax.md, samples_errors_at_sigma_instance_prior_valued_model.md, ep_hierarchical_scatter_moment_matching.md). The Cortex has no home for an umbrella ledger either: phases/ requires Project: to equal a projects.yaml key plus a run State:, and there is no graphical_ep project row. So the Mind graphical-ep entry keeps its ledger:, and instead the Cortex epics.md entry's stale ledger: PyAutoMind/research/graphical_ep/ep_campaign.md (missing draft/) is corrected.

Key Files

• PyAutoMind/epics.md, queue.md, REFERENCE.md, ROUTING.md, AGENTS.md, scripts/status.sh, scripts/spawn.py, docs/pyautobrain/spawn_spec.md
• PyAutoBrain/AUTONOMY.md, agents/faculties/sizing/_sizing.py, agents/conductors/intake/_intake.py, agents/conductors/feature/_feature.py + the three taxonomy docs
• PyAutoCortex/phases/{slope_hierarchy,inference_programme}/, epics.md, batches/2026-08-31-am.md

Testing

Mind pytest -q + python3 scripts/lifecycle.py check; Brain pytest -q (sizing/intake/batch at minimum); Cortex pytest -q + python3 scripts/cortex.py check; pyauto-brain batch plan --dry-run still runs; TZ=UTC PyAutoBrain/bin/pyauto-brain intake --apply dashboard regenerates the Mind dashboard.

Delete the science-epic residue the Cortex split left in Mind

Type: maintenance
Target: pyautomind
Repos:

• PyAutoMind
• PyAutoCortex
  Themes:
• mind-workflow
  Difficulty: medium
  Autonomy: safe
  Priority: high
  Status: draft
  Consequence: judge
  Witness: grep -rn "cortex-half\|epic-slice\|theme-sweep\|^Lane:\|experiment/" PyAutoMind --include=*.md --include=*.sh --include=*.py returns only complete/ history; epics.md has no jax-inference-profiling entry; the five named RAL prompts exist as PyAutoCortex/phases/** files with State: planned and are gone from draft/; batches/2026-08-31-am.* is gone from Mind and present in Cortex; Mind pytest -q and lifecycle.py check pass; pyauto-brain batch plan --dry-run still runs
  Review-minutes: 20
  Unattended: ready
  Epic: mind-post-cortex
  Phase: 2
  Filed: 2026-09-03

Phase 2 of mind-post-cortex. Touches REFERENCE.md, ROUTING.md,
scripts/status.sh and possibly tests/ → this is a code-side Mind PR (the
ledger auto-merge will not take it); the Cortex side is its own PR.

The residue (all verified 2026-09-03)

1. cortex-half: keys in epics.md (entries jax-inference-profiling,
   cluster-strong-lensing, graphical-ep, euclid-dr1-prep). Read by no
   code: _intake.py _EPIC_FIELDS = ("title","ledger","status","notes").
   Delete the key everywhere; where the Mind half is live (cluster-strong- lensing, graphical-ep, euclid-dr1-prep) fold one clause into notes:
   ("science half: PyAutoCortex epics.md#"). Update the epics.md
   schema paragraph if it mentions the key.
2. jax-inference-profiling entry in epics.md: its own status: says
   the whole programme lives in Cortex and slices ship as autolens_profiling
   PRs, not Mind prompts. One member prompt remains
   (draft/feature/workspaces/phase_5_dev_leg_prepare_the_mesh.md) — detach
   it (drop Epic:/Phase:, keep as an ordinary draft). Retire the entry
   (status COMPLETE — moved to Cortex 2026-09-01, then the --retire
   move; the ledger is external so only the entry text is archived).
3. Lane: header (REFERENCE.md "Lane" section, queue.md schema lines,
   batches/AGENTS.md if mentioned). One draft uses it. Remove the Mind-side
   documentation and the one header
   (draft/maintenance/autolens_profiling/legacy_point_output_sweep.md, which
   moves to Cortex anyway — see 6). Leave Brain's _batch.py lane filter
   alone: Cortex members are Lane: local-dev by construction and the Cortex
   REFERENCE owns that key now — add one line there if it is not already
   documented.
4. kind: epic-slice / kind: theme-sweep in queue.md (schema + the
   "Three kinds" paragraph) and REFERENCE.md: zero consumers
   (_batch.py globs draft/**/*.md only). Reduce the schema to kind: prompt | retired; drop the lane: line. Sweep the four kind: retired
   entries into a short "Retired" footer or delete them.
5. experiment/ work-type: no folder on disk, 0 drafts, 5 historical
   records. Remove it from ROUTING.md's table, AGENTS.md's work-type
   list, README.md "Prompt taxonomy", scripts/status.sh, and Brain's
   _sizing.py work-type table / autonomy cap ("experiment": ... at the
   ~L55 and ~L936 dicts) plus any Brain test fixture that enumerates
   work-types. Existing complete/2026/** records keep their folders.
   Leave the research cap alone (verdict-shaped research parks for a
   human by design).
6. Five run-shaped prompts → Cortex phases. draft/research/graphical_ep/ ep_campaign.md and slope_hierarchy_methods_writeup.md,
   draft/research/autolens_profiling/cluster_gradient_search_benchmark.md
   and multiband_compile_census_completion.md,
   draft/maintenance/autolens_profiling/legacy_point_output_sweep.md. Read
   PyAutoCortex/REFERENCE.md (phase-file schema, State:, Gates:) and
   docs/schema_decisions.md; write each as phases/<project>/<slug>.md
   with State: planned under the matching project in projects.yaml
   (graphical_ep → slope_hierarchy / ic50_workspace as fits; profiling
   ones → inference_programme). Keep the original prose as the phase body.
   Delete the Mind drafts (git rm) and, if ep_campaign.md is the
   graphical-ep epic ledger, repoint that epics.md entry's ledger: at
   the Cortex file or retire the Mind entry if no Mind members remain.
7. batches/2026-08-31-am.md + packets/2026-08-31-am.html +
   reviews/2026-08-31-am.md: all-science; the cortex birth migration map
   (complete/archive/epics/cortex_birth_epic.md) says they move to the
   Cortex, and PyAutoCortex/batches/ already holds a 2026-08-31-am.md.
   Diff the two; if the Cortex copy is complete, git rm the Mind trio,
   else port the missing content first. Fix any batches/AGENTS.md or
   status-box reference that assumes the am slot exists in Mind.

Then: pyauto-brain intake --apply dashboard, Mind pytest -q,
lifecycle.py check, Cortex pytest -q (its ledger_merge / phase parsers),
open both PRs.