ROADMAP.md lags v13.2 by four weeks: nine dead anchors, missing scope

Context

Operator, 2026-09-23 ~00:40 CEST, after handing the whole v13.2 problem scope to seven peers: "ROADMAP.md did describe the release scope, but got stale. I guess not even the repo split is in there. Creating a ticket to update it, adding all missing items with priorities might help us — future sessions or compactions especially." The roadmap is the declared session intake (its own last section says so), so its staleness is paid on every boot and every compaction recovery.

Operator, ~01:00 CEST, second input: with several repositories we are no longer bound to an org release; FM's release requires an outsider to get their own Agent OS running, which requires a first-run wizard; several releases mean several release notes — one set "would be a book". Measured per-repo release state and its consequences: D#19050 comment, 2026-09-22.

The Problem

ROADMAP.md was last edited 2026-08-28 (36c463ba5a, #17830). Measured on dev@e66b6f8142 tonight:

Nine of its issue anchors no longer resolve in this repository. They were transferred out with the split, and the roadmap still links them as neomjs/neo issues: #14560 → neomjs/neo-agent-institution#10, #13448 → neomjs/neo-agent-institution#9, #14472 → neomjs/neo-agent-brain#122, #15798 → neomjs/neo-agent-brain#90, #14442 → neomjs/neo-agent-brain#123; #14565, #14570, #14304, #12679 also fail to resolve (destinations not checked). The GraphQL lookup answers "Could not resolve to an issue"; the web redirect hides this from a human reader, the API does not hide it from an agent.

Three cornerstone anchors are closed and one graduated: #13158 (Qt-parity docking polish) closed; #14581 (convergence-weighted Golden Path) closed; #12456 is already noted as done in prose; D#17818 (the DockLayouts rewrite ideation) closed 2026-08-29 and its implementation epic #18303 carries no milestone.

The org picture is wrong. The text says "four repos": devindex (its own workspace since the split), github-content-sync (the corpus repository, publishing hourly since 2026-09-20) and pages (the deployment target) are absent, and D#17247 (the six-repository split) closed 2026-09-21.

The release model is wrong. "v13.2 is the first org-spanning release" (#17830) couples the fastest-moving product (the Institution / FM, 0.1.0, pinning a dev engine commit) to the slowest gate. Only the engine has ever cut a release (13.1.0); skills already ships on its own npm cadence (0.1.14); the Brain's release form is the image cut (neomjs/neo-agent-brain#253). Per-repo cadence is the de-facto state.

This week's release-gating work is absent: deployment topology (#19047, D#19050) and content scope (D#19051); the corpus consumer cutovers under #17416 and the engine-side retirement of resources/content (17,929 tracked files, 193 MB); KB currency (neomjs/neo-agent-brain#402 ← #237 ← #253, the operator's top item because it re-enables the Dream Pipeline and the Golden Path); the flagship film and the Workstation tour refresh (#15252); the Hacktoberfest door (#18985, opens 2026-10-01); the consumer-install blocker #19052. Milestone #9 reads 104 closed / 9 open, and #18303, #17416, #15252, #18985 are not on it.

Stewards drifted: the text names @neo-fable as steward of #15798 and of cornerstone 2's rewrite path; #15798 left the repository and the rewrite path is #18303 (@neo-gpt).

The Architectural Reality

• ROADMAP.md is next-release-only by rule (.agents/skills/update-roadmap/references/update-roadmap-workflow.md): cornerstones + rationale, an explicit deferred set, one named self-selected steward per cornerstone; replace, never append; no exhaustive item list ("stales → FAIL"). The rule is per repository: each product repository holds its own next release.
• Anchors are prose links; nothing checks that they resolve. #15217 (2026-07-16) fixed the same drift class by hand, and its author declined a freshness lint on one incident. This is the recurrence.
• The "Shipped history" pointer targets resources/content/release-notes/, which D#19051 may move.

The Fix

One documentation PR that replaces the v13.2 section in place:

1. Thesis. The engine's 13.2.0 gate: its site deploys and its notes ship. Retire "the first org-spanning release"; name the three release lines — engine 13.2.0, the Brain runtime images, the Institution / FM (D#18965's first-run path) — each on its own cadence with its own notes, and the Knowledge Base prerequisite (neomjs/neo-agent-brain#253 → #237 → #411) as the local plane's, not a line's gate. (Corrected 2026-09-23 after review; the first draft ordered the lines engine → Brain → Institution and put "against a published engine" on the FM path, both derived rather than verified.) Deployment as part of the engine's gate satisfies #19047 AC-5 by construction; coordinate with @neo-opus-ada so the row is written once.
2. Cornerstones, priority-ordered as the operator steered tonight, each with its live anchors and its live steward (read from the assignee, never pre-assigned; unstewarded rows say so):
   1. KB currency — neomjs/neo-agent-brain#402 / #237 / #253
   2. Deploy topology + content scope — #19047 (D#19050, D#19051), #19052
   3. Corpus consumers + engine mirror retirement — #17416 (neomjs/neo-agent-brain#246 for the FM feed)
   4. Docking, Workstation and the flagship film — #18303, #18680, #15252 (+ #16498, #17871)
   5. Contributor door — #18985 (dated)
   6. Release notes + launch — #14800 (engine notes), #14790
   7. FM cockpit + local onboarding — #13015, #14230, D#18965 (the FM line's gate, pointed at, not owned here)
3. Org table: engine, brain, institution, skills, devindex, github-content-sync, pages — one line each: what the engine's release means there, and where that repository's own next-release section lives (or "none yet").
4. Deferred set: keep it, re-verify each anchor resolves, and place #15490 / #14781 in or out explicitly.
5. Anchors: every #N resolves in the repository it names; cross-repo anchors are written as owner/repo#N.
6. Milestone: put the release-gating epics on milestone #9 (#18303, #17416, #15252, #18985), or record why not.

Acceptance Criteria

• AC-1 Every issue and discussion link in ROADMAP.md resolves in the repository it names; transferred anchors point at their new home.
• AC-2 The cornerstone table is priority-ordered with one live steward each or "unstewarded"; an anchor that closed while the section was written stays only as a marked completed prerequisite. Restated 2026-09-23 after review: "only open work" was false the moment #18303, #19052 and Brain #413 closed under it.
• AC-3 Deployment (#19047) and KB currency (neomjs/neo-agent-brain#402 / #253) appear as release scope; #19047 AC-5 is satisfied by this PR or explicitly by Ada's.
• AC-4 The org paragraph lists all seven repositories with the engine release's meaning per repository, and states where each line's next release stands — a real pointer, or that none is declared yet (checked: no sibling repository keeps a roadmap or next-release section on 2026-09-23).
• AC-5 The deferred set is explicit and every entry resolves.
• AC-6 Milestone #9 carries the release-gating epics named above, or the roadmap says why not.
• AC-7 The section is a replacement, not an appended layer, with no shipped-history or vision restatement (the skill's MUST NOT list).
• AC-8 The roadmap names the three release lines, each on its own cadence with its own notes, and keeps the Knowledge Base prerequisite on the local plane rather than inside any line's gate; "the first org-spanning release" is gone. Reviewed cross-family, since it records a release-model decision. Restated 2026-09-23 after review: the earlier "their order" (engine → Brain images → Institution) derived a sequencing from the existence of a dev pin and put the engine gate's own KB prerequisite behind the engine; the lines are independent, the prerequisite is internal.
• AC-9 (post-merge) A compaction recovery that starts from ROADMAP.md alone reaches the live epics, checked by a peer booting fresh.

Out of Scope

• Deciding the deployment topology or the portal's content scope (D#19050 / D#19051).
• The FM first-run wizard's design (D#18965) and the Brain's image cut; this ticket only points at them as the other lines' gates.
• Splitting #14800 into per-repo notes epics; that is the #14800 owner's call, recorded there.
• Moving release notes or the "Shipped history" target (D#19051).
• A roadmap-anchor freshness lint. Optional follow-up now that the drift recurred; file it only if the team wants it.

Avoided Traps

• An exhaustive item list: the roadmap names cornerstones; the milestone query is the full set.
• Pre-assigning stewards: live assignees are facts; unstewarded rows stay visibly unstewarded.
• Fixing anchors in prose only: cross-repo anchors need the full form, because a bare #N means this repository.
• One org roadmap that restates every repository's scope: each repository owns its next release; the engine's roadmap points, it does not copy.

Related

#19047 (AC-5) · D#19050 (cadence input posted there) · D#19051 · D#18965 · #17416 · #15217 (previous roadmap sync) · #17830 (last roadmap edit) · neomjs/neo-agent-brain#402 / #253 · #15252 · #18985 · #14800 · #14790 · #13015 · #14230 · #18303 · #19052

Sweeps: live latest-20 open at 2026-09-22T22:24:59Z — no roadmap ticket (newest: #19052); A2A claims 21:58Z–22:22Z — none on the roadmap; Memory Core: #15217's author declined a freshness lint on 2026-07-16 pending a recurrence; own-assignment: #15252 only. Structure-map gate: N/A (documentation, no ai/ or .mjs placement).

Origin Session ID: 28a95670-a9dc-4214-8387-189e996c16b6 Retrieval Hint: query_raw_memories("ROADMAP.md stale anchors transferred out of the engine v13.2 refresh priorities per-repo release lines")

Ownership: self-assigned; text-only lane, PR after D#19050 / D#19051 close their first divergence window.

🪢 Mnemosyne (Claude Fable 5.1 · Claude Code) · session 28a95670-a9dc-4214-8387-189e996c16b6