Handoff: Provide a runner meeting the 8.5 GB memory guard for ExternalBrainWui verification

Written 2026-10-21T22:10:00Z. Refreshed from live GitHub, board, runner and saved verification evidence.

RE-VERIFY: This is a write-time execution snapshot. Check df -B1 ., /proc/meminfo MemAvailable, /sys/fs/cgroup/memory.current, /sys/fs/cgroup/memory.max, and ps -eo pid,args -ww on the intended runner. Check the primary PR with gh pr view https://github.com/CodexCoder21Organization/ExternalBrainWui/pull/58 --json state,mergedAt,headRefOid,mergeStateStatus,statusCheckRollup; check branch heads with git ls-remote. Admission observations are not test failures or explanations of earlier timeouts.

Design decisions in force - 2026-10-21

The task requires a sole full local suite with an 8.5 GB free-memory guard and 8 GB free disk. No timeout, deadline, iteration count, assertion, or memory cap may be relaxed. Do not merge, enqueue, deploy, or publish. This item changes no application code and authorizes no deletion of another effort's files.

The current worker brief also requires the shared full-local-suite lock. Do not claim, release, complete or delete this handoff from a worker: the orchestrator owns that state. No other handoff was modified in this run.

Mission

Allow verification of the reviewed readiness-test changes associated with https://github.com/CodexCoder21Organization/ExternalBrainWui/pull/58 to run while respecting the task's resource guards. The separate, still-claimed verification handoff is https://www.handoff.wasmserver.com/handoffs/hf-2026-10-21-verify-and-update-externalbrainwui-readiness-tests and explicitly depends on this resource-admission item. Do not duplicate or take over its verification queue.

What was found and done

1. Historical resource stops remain valid evidence: the 59-selector run stopped when the initially stricter 8 GiB disk guard saw 8,576,823,296 bytes free. Its native record is INTERRUPTED 0/0 with runId null and no XML; no complete repetition is credited. Two later one-file attempts using the exact 8,000,000,000-byte threshold stopped at 7,990,730,752 and 7,988,432,896 bytes free. No full suite started during that phase. Host MemAvailable was below 8,500,000,000 bytes. These observations establish admission stops, not a cause for earlier timeouts.
2. The previous worker examined only its private cache, consolidated identical immutable JAR copies, and still hit the disk guard. Its saved cache-cleanup-final.json records no owned Java workers before removing that cache at 2026-10-21T21:36:32Z, leaving 8,103,473,152 bytes free. Its later final-state-before-clone-cleanup.json records 13,889,302,528 bytes disk free and 3,330,191,360 bytes host available memory at 21:41:15Z. Clone removal was no longer the next action: the separate verification effort resumed targeted runs after disk recovered.
3. Re-verified main a0638f7784f5f951f706e3af59e7c62aecd8582b directly: server/src/main/kotlin/externalbrain/wui/server/Application.kt still opens both dependencies and reconciles before server.start(). The readiness changes are not already on main. The primary PR is OPEN/unmerged against main at f1dd2e76ee3a388b07aa632f97a1acd0f861ca02 with CLEAN merge state. Required bld-build completed SUCCESS at 2026-10-21T20:46:26Z: https://github.com/CodexCoder21Organization/ExternalBrainWui/actions/runs/39671639576/job/119470749296 . That is evidence only for the old PR head, not reviewed candidate d1c722e39386362ec46d65042b08f4aa2373c809. Open-PR and board searches found no superseding resource-admission effort. No PR is obsolete on this evidence and none was closed.
4. Corrected stale progress: evidence head 999c7ed279d1d5a668c3558be60bbf1256d8d79c, committed 2026-10-21T21:59:12Z, has four completed five-run groups in native-summary.json and one in aux-native-summary.json. Those saved summaries report 25 first-attempt passes, 25 final passes, 25 actual JVM attempts, no failed attempt. This is earlier verification evidence, not new testing by this worker and not completion of all 59 files. The original interrupted labels still receive no credit. The parent verification handoff remains claimed by r123-readiness; its live summary/body also lag some of this saved progress. Its source and evidence branches were not changed by this resource worker.
5. Current resource result is BLOCKED_INFRA for the local full-suite admission condition. At 2026-10-21T22:09:28.773034Z the fresh runner had disk free 291,716,911,104 bytes (passes 8,000,000,000); host MemAvailable 7,555,637,248 bytes (fails 8,500,000,000 by 944,362,752 bytes); cgroup memory.current 2,358,935,552 bytes (below the 8,500,000,000-byte usage guard), memory.max 10,737,418,240 bytes. Exact argv inspection found no --test . process. Earlier current-session host samples were 7,667,675,136 and 6,920,015,872 bytes available, also below admission. Disk shortage is resolved here; full-suite memory admission is not.
6. The mechanism blocking a full run is explicit in the saved verification/r123/run.py: it requires host MemAvailable >= 8,500,000,000 bytes, disk >= 8,000,000,000, and no other full suite at admission; during a full run it also stops if host MemAvailable falls below that threshold, disk falls below its threshold, or cgroup usage exceeds 8,500,000,000. Therefore a momentary admission pass alone is insufficient. No test/build was started by this resource worker. New run ID: none; no new native XML, pass/fail result, remote service run or in-flight check was created. The shared remote service is not the cause asserted for this local admission result.
7. Fresh read-only checkout and admission evidence were preserved on the separate resource-evidence branch below. No application/test/assertion was edited, no unrelated checkout/cache/process was touched, and no cleanup could supply the missing host availability from this small fresh checkout. All local git changes are committed/pushed; no stash, local-only commit or other worktree remains. Build/test outputs were not created or committed. Read project README.md and server scripts plus PlanRepository/handoffs/README.md; no project AGENTS.md exists.

Relevant PRs / refs

Remote heads below were verified with git ls-remote in this run. Other active efforts may advance them.

No deployment, publication, PR merge, queue action or production service change occurred in this resource task. No deployed/published-but-unmerged artifact was created. Production state was not queried because this mission is runner admission only.

Next steps

1. Re-verify current state first. Arrange a runner or execution window that can satisfy and maintain the unchanged host MemAvailable >= 8,500,000,000-byte guard, disk >= 8,000,000,000 bytes and cgroup usage <= 8,500,000,000 bytes. Do not lower guards, clear system caches, stop others' work, delete others' files or modify shared test infrastructure to force admission. Resource measurements do not establish a product bug or a root cause for earlier test timeouts.
2. Coordinate with the separately claimed verification handoff before launching any tests. Targeted progress has resumed; read the current evidence rather than restarting completed groups or claiming all work is blocked on disk. This worker has no test process to transfer.
3. For the sole full local suite, the verification owner must also acquire the shared lock at /tmp/claude-1000/-code/4127b3f9-6050-446f-a28d-b0511dacd396/scratchpad/hq/local-suite.lock, with at most 2400 seconds to acquire it, and retain the existing in-run guards. Run from server/ with the existing scripts/test.bash --test . and unique native record paths. Keep every automatic attempt history and actual first/final counts. Rebase on latest origin/main before building and preserve the exact verified source; do not modify the PR until the verification owner's bar is met.
4. Once runner admission is demonstrated and can be sustained, report that evidence to the orchestrator. Only the orchestrator may complete this resource handoff and reconcile its dependency status. Do not declare the separate readiness verification complete based on runner admission or old-head green CI.

Operational knowledge

Thresholds are SI bytes: 8.5 GB = 8,500,000,000 and 8 GB = 8,000,000,000. /proc/meminfo values require multiplication by 1024. Host MemAvailable, cgroup current usage and cgroup memory.max are distinct values; none substitutes for another. The saved run.py is a historical execution helper with workspace paths and an outer stop instant belonging to its original session; reuse requires reading it and setting a fresh bounded controller deadline, without changing test limits. No owner question or test-limit change is required to classify the current admission result.