docs(mt#1505): Roadmap stall audit synthesis + structural-process retrospective

[minsky-ai]

Summary

Synthesis writeup for mt#1505, the umbrella roadmap stall audit. Mirrors the sibling reports at docs/research/mt1539-workaround-cluster-audit.md (Child A) and docs/research/mt1540-lynchpin-stall-audit.md (Child B). Companion to the Notion writeup at https://www.notion.so/35a937f03cb481119a08f3c47bd8e6e7.

Key Changes

• New file docs/research/mt1505-synthesis.md. No code changes.

Outputs (operational, not in this diff)

• 8 stalled clusters identified across both diagnostic shapes (5 Shape A workaround-firing × stall-age, 3 Shape B lynchpin × inactivity).
• 6 new MegaParent rollups filed by children during the audit: mt#1563 (session liveness), mt#1564 (auto-mode skill chaining), mt#1565 (TOCTOU enumeration), mt#1570 (ADR-008 transport bindings), mt#1571 (ADR-007 retrofit), mt#1572 (T0 progressive adoption).
• mt#1035 anchor case (System 3* detector design DONE without implementation children) closed via mt#1541 + mt#1543 filings during mt#1508 plan-task.

Recommends shipping a parent-rollup-completion check (Mechanism #7, filed as mt#1684) as the highest-leverage next structural fix. The hook on tasks_status_set blocks DONE on a parent task with non-DONE children — directly addresses the mt#1503 failure mode (set DONE 2026-05-04 with 0/7 children DONE; reopened 2026-05-07).

Bridge mechanism #1 (monthly cron audit re-run) recommended until mt#1541 + mt#1543 land and structurally subsume the temporary-mechanism-budget bridge.

Side effects (already executed)

• mt#1684 filed (parent-rollup-completion check)
• DB memory e81315d4-... (feedback_temporary_mechanism_budget) updated with audit findings
• mt#1451 spec amended with rollup-pattern observations as input for mt#1535 convention work
• Notion synthesis writeup posted under Minsky parent

Testing

• Documentation-only PR; no functional changes.
• Sibling reports (mt1539-*.md, mt1540-*.md) followed the same format and merged previously without test gates.

[minsky-reviewer]

Independent adversarial review (Chinese-wall)
Reviewer: minsky-reviewer[bot] via openai:gpt-5
Tier: unknown

---

Docs-only PR adding a synthesis writeup for mt#1505. The document is coherent and aligns with the already-merged Child A/B audits, with no functional code changes. I flagged several non-blocking nits to improve precision and durability: clarify the mt#1503 child-status phrasing, add a durable reference or excerpt for the Notion link, align the tasks_status_set hook naming with in-repo conventions, include explicit ADR filenames for ADR-007/008, and separate the anchor-case row from the stalled-cluster table to avoid conflation. No blocking issues; ready to merge after optional cleanups.

Findings

• [NON-BLOCKING] docs/research/mt1505-synthesis.md:22 — Ambiguity in child-status summary for mt#1503 row (CLOSED child counted among “active children”)
  In the combined stalled-cluster table, the mt#1503 row lists children mt#1073, mt#1065 (CLOSED), mt#1405, mt#1477 and the status cell reads 0/4 active children done. Since one child is already CLOSED, the phrasing "0/4 active children" is ambiguous — it implies all 4 are active when one is not. Consider clarifying to something like 0/3 active children done (1 CLOSED) or adjusting the denominator to reflect only non-terminal children to avoid misinterpretation.
• [NON-BLOCKING] docs/research/mt1505-synthesis.md:8 — External Notion link lacks archive or content digest; consider adding durable reference
  The header references a Notion URL (https://www.notion.so/35a937f03cb481119a08f3c47bd8e6e7). For long-term reproducibility, consider including a brief excerpt, exported PDF in-repo, or a content hash/date so readers can verify the synthesis source even if the external doc moves or permissions change.
• [NON-BLOCKING] docs/research/mt1505-synthesis.md:92 — Out-of-repo paths and DB memory IDs cited without in-repo context; add verification note or appendix
  The "Actions taken" and "Cross-references" sections reference DB memory IDs (e.g., e81315d4-6241-4e5b-9118-f44398ae52a5) and other out-of-repo artifacts. Per repo readers cannot verify these. Consider adding an appendix with minimal context or a NEEDS VERIFICATION note indicating these are external system references, or link to any in-repo mirrors that track these updates.
• [NON-BLOCKING] docs/research/mt1505-synthesis.md:55 — Mechanism #7 reference to tasks_status_set hook assumes naming/API; suggest aligning with in-repo conventions or citing file
  The text suggests a hook on tasks_status_set and cites similarity to existing hooks like parallel-work-guard.ts, check-branch-fresh.ts, block-subagent-bypass-merge.ts. To aid implementers, consider linking to the actual in-repo hook filenames/paths or clarifying the canonical event name to avoid drift if the event is named differently (e.g., task_status_set).
• [NON-BLOCKING] docs/research/mt1505-synthesis.md:36 — Inconsistent naming: “CognitionProvider” vs ADR-007 title and subsystem names
  The cross-shape pattern 1 references "ADR-007 (CognitionProvider)" and ADR-008 "attention-allocation"; ensure these labels match the exact ADR titles in docs/architecture/adr-007-*.md and adr-008-*.md for searchability. A small parenthetical with the full ADR filenames would improve traceability.

Inline comments

• docs/research/mt1505-synthesis.md:28 — Minor: the table header says "Stalled-cluster table (combined)", but the final row is labeled "A anchor" and "(no rollup)". Consider marking this row as an "Anchor case" separately or adding a footnote so readers don't conflate it with the stalled clusters proper.