concepts: declare the three edges PR #219 dropped to the body cap

[sameera]

Hand curation of the concept store

PR #219 (distillation of epic #185) recorded three real interactions of backlog-stub as prose rather than as declared touches edges, and flagged the loss in reviewer note 1. Each neighbour sat at 398–400 body words against the old 400-word cap; a reciprocal Integration Points bullet costs ~15 words, and compressing a neighbour's still-true content to make room is not a legal move. So the drain dropped the edges.

Epic #220 re-cut the cap to measure a page's own content and to bound the neighbour list per entry instead. The reciprocity fan-out can no longer fail, which is exactly the success metric #220 stated for these three pages. This PR declares the edges.

Edges declared

Neighbour	Interaction
publishing-config-resolution	supplies the unplanned label and the epic classification the batch filing path stamps on a stub — resolved keys, never hard-coded
issue-sourced-planning	its resolver refuses an unplanned epic by name (epic-not-planned) rather than emitting one whose story set is empty
durable-close-record	the close comment carries the deferred-scope numbers and the backlog query, so stubs are filed after the checkpoint and before the comment is composed

Each verified against source before writing: .claude/skills/nxs-epic-resolve/SKILL.md (the unplanned-epic refusal), delivery_config.py (unplanned-label as a resolver key), .claude/commands/nxs.close.md (step 3b filing and the Deferred-scope comment lines).

Every edge is bidirectional: backlog-stub gains three bullets and one Decision Log entry; each neighbour gains backlog-stub in touches, one mirrored bullet, and one Reciprocal link from backlog-stub entry.

Why hand curation, not a drain

The distiller is the store's single producer, and the README's one carve-out is deliberate manual curation marked last_updated_by: "manual". Both epics involved had already drained — #185 through PR #219, #220 through PR #236 — so no queue entry was left to carry these edges. The alternative, waiting for some future drain that happens to touch one of the four pages, leaves blast-radius retrieval on backlog-stub wrong for an unbounded time, which is the silent degradation note 1 was raised about. That trade-off is recorded as the refuted alternative in the new Decision Log entry.

Prior art for the manual source on a reciprocal entry: decision-record.md, committed-queue.md, nexus-pipeline.md.

Checks

• pnpm nexus:validate-concepts — 0 blocking, 5 advisories, all pre-existing on other pages. No advisory on the four touched pages; every new bullet is under the 25-word advisory.
• Append-only pass (--base main, the four changed pages) — OK. One new Decision Log entry per page, prior entries untouched.
• pnpm nexus:check-atlas — OK. docs/concepts.md regenerated (ordering shift only, no count change).
• Anchors untouched: no code changed in this PR, so there is nothing to re-derive or re-stamp.

Scope

Only reviewer note 1 of PR #219. Note 2 (store-wide cap saturation) is what #220 fixed; notes 3 and 4 were resolved within #219 itself.

🤖 Generated with Claude Code