plan: #3075

[PolyphonyRequiem]

---

requests_parent_change: false
ancestor_plan_generations:
root: 1

Plan for #3075 (root #3071)

Promotes plan/3071-3075 into plan/3071.

AB#3075

[PolyphonyRequiem]

🤖 Polyphony plan reviewer (advisory)

Summary: Solid, well-scoped plan for the archivist + first cross-platform write. Motivation, decomposition into three tasks, and dependency call-outs to #3072/#3073 are clear and defensible. Main concerns are around under-specification of the decision schema/citation shape, a soft acceptance criterion on the cross-platform proof, and a contingency in the dependency section that could become a scope-leak if #3073 slips.

Strengths

• Crisp problem framing: ties scratch-curation gap directly to parent #3071 and downstream #3077 needs (Obsidian/Johnny-Decimal, citations, freshness).
• Clean separation of agent vs deterministic code (archivist agent emits decisions; promotion writer is non-agent and consumes them) — keeps the LLM boundary narrow and testable.
• Honest dependency analysis on #3072/#3073 with a fallback for #3073 ordering.
• Cross-platform proof is anchored here by design, not deferred — and prefers the harness over live CI, which matches repo testing posture.
• Acceptance criteria include both behavioural shape (decision per artifact, citation fields) and the build/test baseline.

Concerns

• Decision schema is named but not specified. "JSON-schema-validated" + the five relevance axes are mentioned but the schema (fields, enums, required vs optional, where it lives) isn't pinned. Downstream tasks and the reviewer of task-1 will have to invent it.
• Citation/front-matter shape is under-specified. source_url, capture_date, freshness band are listed, but front-matter-vs-sidecar is left as an "or", and the freshness band values aren't enumerated. This is the artifact format #3077 will consume — pinning it here avoids rework.
• Cross-platform combo is "tentative". AC #4 says the chosen combo is recorded in the plan, but the plan currently says "Tentative: source-on-GitHub + research-on-ADO". Either commit, or make the AC explicit that task-3 selects and records before implementation starts.
• expand semantics handoff to #3076. Plan says expand is recorded in the run artifact for #3076's loop-back. The shape of that record (what #3076 will consume) isn't specified — risks a coupling break when #3076 lands.
• Fallback for #3073 ordering risks scope drift. "May need a small follow-up" is the kind of phrase that becomes a forgotten Issue. Recommend either: (a) hard-gate on #3073, or (b) explicitly carve the wiring follow-up as a known task-4 stub.
• No mention of state-name sourcing. Not necessarily a violation here (this Issue isn't obviously transitioning work-item state), but if any node touches work-item state, state names must come from polyphony validate / process-config.yaml per the canonical pattern in scripts/scope-closer.ps1:54-60. Worth a one-line note that this Issue does not change state-transition code, to make the reviewer's job trivial.

Suggested changes

• Inline the archivist decision JSON schema (or link a concrete file path the schema will live at) and enumerate the five relevance-signal fields with types.
• Pin the citation metadata shape: front-matter vs sidecar, exact field names, freshness band enum (e.g., fresh | recent | stale | unknown).
• Commit to the cross-platform combo (source-on-GitHub + research-on-ADO) in the plan body, and reference the harness scenario id task-3 will create.
• Add one sentence on the expand record shape so #3076 has a contract to bind to.
• Either hard-depend on #3073 or add an explicit task-4 placeholder for the wiring follow-up so it can't be forgotten.

This is an advisory comment from the polyphony plan_reviewer agent. It does not vote — human review is the merge gate.

[PolyphonyRequiem]

polyphony:approve