[Jammy2211]

The Cortex half of mind-post-cortex phase 2 — issue PyAutoLabs/PyAutoMind#390. Siblings: PyAutoMind#391, PyAutoBrain#343.

The Mind is the software/PR loop now, so run-shaped prompts and the all-science batch slot come here.

Four new phases

Transcribed from their Mind prompts with the original prose kept as the phase body and Migrated-from: naming the source. All State: planned; cortex.py check passes.

File	Phase	Project row	Note
phases/slope_hierarchy/methods_writeup.md	2	slope_hierarchy (dormant)	the NUTS-headline / EP-cautionary write-up of the finished N=5 result. A write-up phase — no runs of its own, so Runs: stays empty and the ruling judges the written claim. Epic: graphical-ep, beside phase 1 n25_scale_up.
phases/inference_programme/cluster_gradient_search_benchmark.md	17	inference_programme	see the caveat below
phases/inference_programme/multiband_compile_census_completion.md	18	inference_programme	the confirmatory census legs left over from the multiband-pyloop-batching ship
phases/inference_programme/legacy_point_output_sweep.md	19	inference_programme	Gates: autolens_profiling#205 — opened 2026-09-01 as this work's gate ref; reuse it, never open a second. A file move, so Runs: stays empty.

projects.yaml was not touched — it is CODE and a human's turn. Every phase landed under an existing project row.

One caveat worth a reviewer's eye: cluster_gradient_search_benchmark was marked ABSORBED 2026-07-31 (same day it was filed) into the Mind prompt draft/feature/autolens/point_source_defaults_campaign.md (phase B, cluster tier). That note is carried verbatim into the phase body, with an explicit "do not dispatch standalone". It is here so the context literals survive the split and so the campaign's cluster tier has a phase to rule against — not as independent work. If you would rather it had been dropped than transcribed, this is the file to say so on.

batches/2026-08-31-am.md

The Mind trio (record, packet, review) is deleted in the sibling PR, so this file is now the record of record. Its notes previously promised "The Mind original stays intact … a verbatim history has to live somewhere live" — corrected, and the three facts the 2026-09-01 transcription had dropped are ported in:

• the pre-framework usage-window-at-dispatch: / heart-ack: / expected-effects: n/a values, now header lines;
• the Mind record's eighth member euclid-phase2-pr46 — a dev retrospective on merged PR chore: canonical organ order — Eyes after Memory, before Heart (PyAutoMind#439) #46, judged structure-ok, 35 review-minutes — named in the notes and still, deliberately, not transcribed as a science member;
• the Mind commit refs behind the packet fix (6491b09c) and the rewind (2fb40b45 / autolens_profiling#200).

Verified before deleting: the packet HTML is byte-identical to the Mind copy, and the Mind review file is a strict subset of the Cortex one.

epics.md

• graphical-ep's ledger: was PyAutoMind/research/graphical_ep/ep_campaign.md — missing its draft/ segment. Fixed. (That file stays in the Mind: it is the campaign map, five Mind drafts still declare Epic: graphical-ep, and there is no graphical_ep project row to hang a phase on. The reasoning is in PyAutoMind#391.)
• jax-inference-profiling is mind-half: none now — its Mind entry was retired in the sibling PR.
• The header prose records that the Mind's reciprocal cortex-half: key is gone and that - mind-half: is the machine-readable half of the join.

Tests

• python3 scripts/cortex.py check — OK
• pytest -q — 134 passed
• dashboard.md/dashboard.html regenerated with canonical PyAutoBrain main and TZ=UTC (32 phases, 22 rulings); dashboard --check reports current

Refs PyAutoLabs/PyAutoMind#390

🤖 Generated with Claude Code

https://claude.ai/code/session_013HsZA1ufn3msgPiDFxEXa6