Issue 12584 triage, 2026-10-22 UTC Main: 89fe509232 (origin/main fetched before creating isolated worktree) Environment: Linux x86_64; Node 24.21.0; real web/server development app. Command: vp run dev (worktree node_modules/.bin on PATH) Worktree: /home/ubuntu/wt/issue-12584 State: isolated .t3; read-only snapshot seeding kept 3 stopped threads and 341 orchestration events. Server port 16440, web port 8400. Paired through collaborative browser. Startup: V2 orchestration recovery completed, terminalizedRuns=0, stoppedSessions=0. Seeding warned of migration IDs 68/69/70 unknown to this main checkout. No live data was modified. No provider agents were launched. Attempt 1: browser loads seeded app and an empty thread. Direct read loop omitted x-t3-orchestration-protocol; 300 requests returned 400. This was an invalid harness attempt, excluded from reproduction claims. A subsequent 1000-request corrected evaluate timed out at 15 seconds; no result was retained, excluded from counts below. Attempt 2: corrected headers x-t3-orchestration-protocol: 2. 200 sequential browser fetches alternating orchestration shell and bounded empty-thread reads. Result: 200 HTTP 200; 1,117,900 response text characters; elapsed 2640 ms. Attempt 3: opened a populated copied thread in the real app. 200 sequential bounded-thread reads, same required protocol header. Result: 200 HTTP 200; 41,571,000 response text characters; elapsed 9306 ms. Server RSS samples from /proc / ps, no forced GC or heap snapshots: uptime 00:49: 552200 KiB uptime 01:26: 643128 KiB uptime 01:41: 629564 KiB uptime 01:55: 627604 KiB uptime 02:11: 627616 KiB No FatalProcessOutOfMemory / heap out of memory / SIGABRT in dev log. RSS is not retained JS heap; these short samples do not establish a leak or its absence. Limitations: not the reporter's Arch SEA/systemd runtime; Node differs from repo-pinned 26.12.1. No sustained provider-session load, 430 live sessions, or 30-48 hour run. This bounded read-load variant did not reproduce the reported crash and cannot establish it is fixed. Dedupe: scanned 938 open issues (50/page through first page older than 90 days), matching heap exhaustion, background service, sustained sessions and FatalProcessOutOfMemory terms; no same-surface duplicate found. Scanned 1382 open PRs; no fix/issue-12584 branch PR or exact fix found. Related 19305 is merged and ancestor of main: native SQLite binding retention, not proof about this V8 abort. Related 23141 and 19585 target issue 18799 unread ACP notifications, not this sustained-service report. Decision: request a current-build reproduction, provider/concurrency/churn details and heap-vs-RSS/retainer summary; leave open.