staged(v2.31.x): Memory artifact ledger - F-48

[jprisant]

Revive the parked plan at docs/internal/release-plans/_unreleased/project-memory/plan_project-memory.md as its own release. The project-memory file doubles as the orchestrator artifact ledger, and its provenance stamping aligns with the maintainer-local innovation X-3 (artifact provenance and the upgrade loop). This workstream integrates the ledger interface by documenting how the orchestrator records execution state, artifact hashes, and provenance chains across skill invocations.

Promotion trigger: v2.31.0 shipped + orchestrator artifact-ledger interface agreed.

Target release: v2.32.0

Links: plan_project-memory.md, orchestrator artifact ledger interface (X-3)

[jprisant]

Researched for v2.32.0 (2026-07-31); disposition pending maintainer ruling. Full findings in the decision-stage plan: docs/internal/release-plans/v2.32.0/plan_v2.32.0.md (candidate C-1, decisions D1/D2/D3).

Summary: the parked project-memory plan + spec remain current (B1 keystone state file, B2 memory-aware cohort, each sized M separately). This issue ADDS ledger semantics the parked schema does not model (orchestrator execution state, artifact hashes, provenance chains), which needs a 1-2 page delta spec before any build. Pre-build gates: ratify parked D1 (structure-over-prose), rule parked D2 (write posture; split out as its own trust decision D3 in the plan), lock parked D3 (cohort membership), reconfirm the F-54 provisional ID. X-03 (artifact provenance) does not depend on this and is not blocked by it.

Recommended: ratifications + delta spec in v2.32.0; the B1 build lands either this cycle (D1-B/D2-B, B1-only converts the effort L to M) or in v2.33.0 after S2 behavior is characterized (D1-A, recommended). Known risk, stated in the plan: this train has slipped twice already.