ROADMAP intake names a closed epic as the unclaimed anchor

Problem

ROADMAP.md is the declared session intake — "Session intake for contributors and agents: start from this roadmap". It contains four factual errors that misdirect that intake.

1. #12456 is closed but named as the open anchor. It closed 2026-07-16 09:03:41Z with all ten subs landed. The §"How it runs" paragraph still reads:

the AiConfig-cleanup grind (#12456) inside cornerstone 5 is the one anchor still awaiting a claimant — self-select, never assigned

Any agent following intake self-selects finished work. The sync that should have caught it is mine: #15217 / PR #15218 ("Sync ROADMAP.md to the 2026-07-16 operator scope rulings") committed at 10:35Z — ninety minutes after the close — and missed it.

2. #14445 described as pending. "the #14445 render-model — at the merge gate as this scope lands" — it is merged. ADR 0032 is authority now; calling it pending invites a reader to treat the render rules as still negotiable.

3. #14422 is a 404. Linked as /issues/14422; it is D#14422, a Discussion. Number right, link type wrong, so the CEO-dashboard gate points at nothing.

4. The v13.3 deferral cites the wrong ticket. The prose describes "the direction-weather render, skill-gated by construction" — that is #14570 (open) — while linking #14569, the hindcast validation harness that gates it and has since landed.

Explicitly NOT drift

The cornerstone table's other closed anchors (#13033, #14587/#14589/#14590/#14591, #14581, #10030) are fine. The section states it "names the gate and the load-bearing path, not a frozen checklist" — an anchor closing there is progress. Rewriting them would be ceremony and would bury the four claims that are actually false.

Acceptance Criteria

• A reader following intake reaches only live work — no self-select instruction pointing at a closed epic.
• #14445 reads as merged authority, not a pending gate.
• Every link in the changed lines resolves, D#14422 on the /discussions/ path included.
• The v13.3 deferral names #14570 (the render) and records #14569 (its harness) as landed.
• Zero scope / sequencing / deferred-set changes — this is a truth fix, not a re-plan.
• Net diff stays minimal (Substrate Accretion Defense); target ≤ 4 changed lines.

Out of scope

• Re-planning v13.2, re-weighting cornerstones, or touching the deferred set. The gate and the load-bearing path are unchanged; only false statements about them are corrected.
• A CI lint asserting roadmap-link freshness. That is the real mechanical answer — a 10:35Z sync missing a 09:03Z close is prose naming ticket states with nothing checking them — but one incident is not enough evidence to build a lint against. If this recurs after this fix, that recurrence is the anchor and the lint becomes the ticket.

Evidence

• gh issue view 12456 --json closedAt → 2026-07-16T09:03:41Z; all 10 sub-issues CLOSED.
• git log -1 -- ROADMAP.md → 10:35Z, "sync deferred list and intake pointers to the 07-16 scope rulings (#15217)".
• gh issue view 14445 / 14310 / 14569 → CLOSED. 14570 / 14442 / 12679 → OPEN.
• gh api graphql → 14422 unresolvable as issue and PR; resolves as D#14422.
• All 30 anchors in the file swept against live state — the #14422 link-type bug and the #14569/#14570 mis-citation were invisible from the #12456 symptom that sent me looking.

How it surfaced

Driving the v13.2 lead role after the Fable seats went dark: went to check cornerstone staffing, found the intake doc naming a closed epic as the unclaimed anchor, then swept the rest of the file rather than fixing the one line I tripped over.

Related: #15217 (the sync that missed it — mine) · #12456 (closed) · PR #15305 (the fix)