Affected Repositories

• PyAutoBrain (primary)
• PyAutoMind

Branch Survey

Suggested branch: feature/autonomous-orchestrators-p2 (both repos); worktree root ~/Code/PyAutoLabs-wt/autonomous-orchestrators-p2/

Implementation Steps

1. agents/faculties/sizing/_sizing.py — LANE_VALUES = ("any","local-dev","ral","gpu","wsl-data"), lane_is_local(), first-valid-declaration-wins, an invalid value is recorded as lane_invalid and read as any.
2. agents/conductors/batch/_batch.py — _lane_ok via lane_is_local; lane_counts in the BatchDecision, printed; other_lane_ready becomes a per-lane breakdown (3 ral, 1 gpu).
3. New agents/conductors/batch/_brief.py + batch brief --check <file>: review-at: ISO and later than dispatched:; integer review-minutes/merge-cap ≥0; lanes ⊆ known values (cloud refused); backlog: none|fill-idle|scoped (scoped needs scoped:); ral-allowance: required with ral; named tasks resolved in draft//active/ (unresolved = warning); merge-cap > 0 refused when every named task is judge and backlog: none (approved refinement); refused while the newest previous record lacks review-minutes-actual:. batch plan --brief <file> restricts the pool to named tasks, then the dip policy, reporting against the brief's review-minutes.
4. PyAutoMind batches/examples/shift_brief_example.md (+ README line); tests/ asserts lifecycle.py never reads a record placed in examples/.
5. agents/conductors/intake/_intake.py — _render_header passes declared Filed/Epic/Phase/Blocked-by/Closes-when after Status: and Lane: after Unattended:; lane inference (RAL/SLURM/sbatch/euclid_jump → ral; CUDA/laptop GPU → gpu; /mnt/c/Workspace/Science → wsl-data; precedence ral > gpu > wsl-data; written only when not any; a declared value wins).
6. Census — by_lane in census(), printed by emit_census, a lane line on the dashboard Backlog section (md + html); lane_invalid as a hygiene flag.
7. Tests for every Witness clause (a)–(e).

Scope note (approved): this fixes the header pass-through half of draft/bug/pyautobrain/intake_classify_file_drops_declared_epic_phase.md; at close-out that prompt is re-scoped to its Memory:-gating half, not closed.

Key Files

• PyAutoBrain/agents/faculties/sizing/_sizing.py
• PyAutoBrain/agents/conductors/batch/_batch.py, new _brief.py
• PyAutoBrain/agents/conductors/intake/_intake.py
• PyAutoMind/batches/examples/shift_brief_example.md, PyAutoMind/tests/

Autonomous orchestrators phase 2: Lane: resource values, the shift-brief validator, intake pass-through and census

Type: feature
Target: PyAutoBrain
Repos:

• PyAutoBrain
• PyAutoMind
  Themes:
• autonomy
• orchestration
  Difficulty: medium
  Autonomy: supervised
  Priority: high
  Status: formalised
  Epic: autonomous-orchestrators
  Phase: 2
  Filed: 2026-11-03
  Consequence: glance
  Witness: pytest PyAutoBrain/tests -q green with new tests that (a) parse Lane: ral, gpu, wsl-data and keep any/local-dev, and reject Lane: any | local-dev; (b) assert _lane_ok("ral","web-github") is False and _lane_ok("ral","local-dev") True for all three new values; (c) batch brief --check accepts PyAutoMind/batches/examples/shift_brief_example.md and refuses the same file with review-at: removed, with lanes: cloud and with a merge-cap: above 0 for a brief whose named tasks are all judge; (d) intake classify on a text naming RAL proposes Lane: ral and a declared Epic:/Phase:/Filed:/Lane: survives --apply unchanged; (e) intake census prints a lane histogram whose total equals the prompt count
  Review-minutes: 5
  Unattended: ready
  Lane: any

Parent: draft/feature/pyautobrain/autonomous_orchestrators_epic.md ("Lane, not Needs").
Starts only after phase 1 merges (epic phase order; the schema this validates is written
there). Phase 1 wrote the words; this phase makes the organism parse them. It is the code half of
the original phase 1, split out because a prose-only autonomy edit is judge and this is
glance with a real witness.

Deliverables

1. Sizing faculty (agents/faculties/sizing/_sizing.py): LANE_VALUES = ("any", "local-dev", "ral", "gpu", "wsl-data"); a helper lane_is_local(lane) that is True for
   everything but any; keep the first-declaration-wins rule.
2. Batch planner (agents/conductors/batch/_batch.py): _lane_ok routes through
   lane_is_local; the BatchDecision prints the lane histogram of the pool; the
   "other lane's ready count" line names the resources (3 ral, 1 gpu).
3. batch brief --check <file>: validate a shift brief against the schema phase 1 fixed
   in PyAutoMind/batches/AGENTS.md; errors name the key. Ship the example brief under
   PyAutoMind/batches/examples/ (ignored by lifecycle.py check's drift scan — add the
   exclusion with a test). plan --brief <file> restricts the pool to the brief's named
   tasks first, then the dip policy, and reports the review-minute total against the brief's
   budget.
4. Intake (agents/conductors/intake/_intake.py): propose Lane: ral / gpu /
   wsl-data when the text names RAL/SLURM/euclid_jump, the laptop GPU/CUDA, or /mnt/c
   / a Science dataset path; pass declared optional headers (Epic:, Phase:, Filed:,
   Blocked-by:, Closes-when:, Lane:) through _render_header unchanged (this is the
   bug draft/bug/pyautobrain/intake_classify_file_drops_declared_epic_phase_filed_headers*.md
   reports; fix it here or there, not both — whichever lands first closes the other).
5. Census: a lane histogram in intake census and the dashboard's backlog page.

Out of scope

The watcher, dispatch and leases (phase 3); any change to merge authority (phase 1 prose
governs; nothing here merges anything).