PRIO-ZERO: Golden Path frozen 18 days (handoff Jun 3) → swarm navigates a stale forecast → mass derailment; #13626 wired the sync task but the freeze persists + orchestrator never idles

Symptom (operator-flagged, PRIO ZERO)

• 8h nightshift, 4 peers online → zero work on the v13.1 harness board (#8) or the orchestrator. The whole swarm derailed into agent-OS meta (stop-hook tuning, contentTrust, blog posts, backlog dispositions).
• resources/content/sandman_handoff.md — the Golden Path the swarm navigates by — last regenerated Jun 3 04:17, 18 days stale. Its "Computed Golden Path" recommends months-old meta work (AGENTS_TENETS authoring, cognitive-load audit, MX-loop discussions; issue-10293 carries structural weight 9.80) because the graph only contains old, meta-heavy nodes. The release/harness/orchestrator tickets have weight 0 — they were never ingested.
• "graph processing running for hours / 2h+ parallel-model processings" with ~zero net-progress (see #13624).

Verified facts (V-B-A)

• Issues are fresh on disk: resources/content/issues/chunk-6/issue-13727.md exists (the hourly data-sync pipeline writes content). The write side works.
• The issue→graph ingestion exists and is correct: IssueIngestor.ingestIssueStates() → GraphService.upsertNode() (IssueIngestor.mjs:139), invoked by SyncService.runFullSync() (SyncService.mjs:232) and the manual npm run ai:sync-github-workflow.
• githubWorkflowSync is a scheduled orchestrator task (taskDefinitions.mjs:201, scheduling/registry.mjs:85, scheduling/pipeline.mjs), interval 30 min (config.template.mjs:375); goldenPath is scheduled at 1 h (goldenPathMs, config.template.mjs:388). Both resolve enabled locally (resolveLocalDeploymentDefault(null) → deploymentMode !== 'cloud' → true).
• So the #13626 fix (the task wiring) is in place — yet the handoff is 18 days stale. The wired task is not effectively running.
• .neo-ai-data/orchestrator-daemon/ is empty (no lease file, no scheduler state) on the github clone.

Root-cause candidates (bisect — needs live orchestrator inspection)

1. Heavy-maintenance lease starvation. githubWorkflowSync, goldenPath, and the session-summary/REM maintenance all serialize through ONE exclusive 6h-TTL lease (HeavyMaintenanceLeaseService, shared across syncGithubWorkflow / syncKnowledgeBase / runSandman / backfill / backup / DreamService). #13624's never-draining session backlog (all-time pending 284 vs 30-day sweep candidates=7) keeps the maintenance loop "busy," so the 30-min sync + 1-h golden-path tasks are perpetually deferred ("Deferred: heavy-maintenance lease held by…", syncGithubWorkflow.mjs:98). This unifies #13624 (no drain / no idle) with the golden-path freeze.
2. Stale orchestrator process (#13289): the running daemon predates the task-wiring; Node doesn't hot-reload → it runs old code without the scheduled task.
3. Stale gitignored config.mjs leaf (config-leaf-false-green class): the live machine's materialized config.mjs lacks the new githubWorkflowSync interval/enable leaves → never scheduled.
4. Golden-path synthesis erroring silently per cycle.

Acceptance Criteria

• Bisect which mechanism(s) freeze the golden path; document the live evidence (lease holder, lastRunAt, process boot SHA, materialized config).
• The Golden Path handoff regenerates on cadence; sandman_handoff.md reflects the current frontier (#137xx), not Jun-3 state.
• Issue→graph ingestion runs so new issues (#10xxx→#137xx, ~3000) are queryable as graph nodes.
• The orchestrator reaches an IDLE state after time X / when the reachable backlog is drained — it must not spin heavy-maintenance indefinitely at zero net-progress. Define the idle condition + bounded-work termination. (Precondition: #13624's net-progress accounting.)
• The shared heavy-maintenance lease cannot let one never-draining task starve the sync/golden-path tasks — fairness across distinct task classes, not just within session sub-tasks.

Links

• Supersedes the intent of #13626 (CLOSED — wired the task, but the freeze persists). Consider reopen vs supersede.
• Root coupling: #13624 (orchestrator runs 14h, backlog doesn't drain — net-progress, not fairness). The idle-requirement + lease-starvation belong to that failure.
• Stale-process candidate: #13289.
• The per-agent enforcement half (stop-hook must inject this Golden Path as the release-ROI direction) is filed separately.

This is the highest-ROI fix in the repo right now: it re-points the entire swarm at the release.