Handoff: Reconcile the changed UrlProtocol head with the stopped validation

Written 2026-10-23T01:54:18.948439+00:00

RE-VERIFY: This is a write-time snapshot. Read the current head/state of https://github.com/CodexCoder21Organization/UrlProtocol/pull/759 before any action. The authoritative remote run record for 38df9d87 is now terminal FAILED; its saved canonical counts and per-attempt evidence are linked below. This lane did not validate the newer PR head.

Design decisions in force

The user authorized one complete suite on the corrected candidate. The binding rules were: “If the PR head moved, stop and report”; “If anything is red, stop and report the failing tests with mechanism; do not push”; and “Never weaken assertions, raise timeouts, or cut iterations; never merge/enqueue/publish/deploy.” The earlier extra-suite approval question is resolved by the C759m authorization and is no longer the blocker.

Mission summary

The owner reported that https://droplets.wasmserver.com/ "is super unreliable, frequently rendering various errors/notices due to bugs in UrlResolver or UrlProtocol or something else" and asked for the underlying root cause to be fixed; the ContainerNursery route stops the WUI after 5 idle minutes, so most human visits are cold starts.

This lane was to validate the corrected combined candidate and update the existing PR only if the complete exact-head suite was first-attempt green and the incoming PR head was unchanged.

What was found and done

1. Fresh owned clone, required README/testing guidance, and incoming backup were verified. The corrected input was rebased onto the then-current main, producing 8184c684307721e59605f9020275f963f1c172cd. The production source remained byte-identical; the rebase brought one upstream test-file change.
2. The six required cases pass on attempt 1 in XML, with no attempt-history rows. Each of the three separate reversal controls is red on all three attempts. Per-registry warning delivery and both incoming error-order corrections remain together; the incoming two error-order test files are unchanged, and the startup scenario retains all nine incoming assertions with the construction-time count restored. No permanent code or test edit was made by C759m.
3. The PR moved from the preserved incoming head to 6d745064e3bd08a3a451afc7403e2682eed6bbc6. It was last verified OPEN. No PR push or description edit was made.
4. One complete remote execution, 38df9d87, ended FAILED at 2026-10-23T01:32:30.930Z. Its canonical result projection has 1833 PASSED, 45 FAILED, 3 CANCELED, and 1329 PENDING/unrun rows (3210 planned). Raw first-numbered completions agree for the 1881 completed/canceled rows, with zero duplicate or conflicting first-numbered records. No complete-suite XML was produced, so there is no XML-based complete-suite pass claim.
5. The controller terminal record reports a provisioning deadline: elapsedMs=610860, limitMs=600000, currentAttempt=11. Its last recorded error is a SandboxException during dropletserviceserver.DropletServiceClientImpl.getDropletCreationStatus. The exact terminal text is preserved in terminal-summary.json and remote-event-summary-terminal.json. Why the call did not complete is not established here.
6. All 45 failed and three canceled canonical identities have saved full diagnostics and source links. Four tests stop at a machine-wide available-memory check requiring 3000 MiB. Two javac helper paths do not resolve in the remote context although the files are locally present and tracked. The causes of the remaining 39 failed assertions/runtime results, the individual canceled rows, and attribution to this candidate versus main are unproven. This validation-only lane ran no baseline or additional suite and made no fix claims.
7. The source/evidence branches were pushed and verified. The raw watcher exited 130 after remote completion. All owned background processes were absent in the local ps check. No merge, enqueue, publication, deployment, production restart, or remote cancellation was requested. Short HardwareControlFabric probes read only this run's saved data; no host files were written.

Relevant PRs and refs

PR: https://github.com/CodexCoder21Organization/UrlProtocol/pull/759 — OPEN at the moved head when last read; re-verify before acting.

No code was deployed or published by this lane. No deployed-but-unmerged row was created. Candidate source and evidence are separate branches; use the candidate commit for code, not the evidence branch source tree.

Primary evidence at 65a52a7d5aee558d1af7611cb1d6b92c0eabb916:

• Final summary and complete failed/canceled diagnostic matrix.
• Focused required XML, plus startup-control.xml, ledger-control.xml, and primary-control.xml.
• Terminal event/state snapshot, findings, and collection limitations.
• Earlier handoff body and references are retained as historical input; its approval blocker and earlier-head suite result do not describe this completed validation.

Next steps

1. Re-verify the current PR head and establish which candidate the owner wants evaluated before another branch update. C759m authorization does not cover replacing the moved head.
2. Review the failed/canceled matrix and the controller provisioning record. Any further tests or fixes require a new scope; this lane's explicit instruction was to stop on red or a moved head. Do not treat the incomplete remote run, the older local suite, or eventual counters as first-attempt green evidence for the current PR.
3. Resolve defects in their owning layer with reliable reproduction under the standard workflow. No timeout increase, iteration reduction, test disabling, automatic merge, publication, or deployment is authorized by this handoff.

Operational knowledge

The initial remote test client reached its unchanged 30-minute wait limit while the run continued; that client exit was not run completion. CLI --attach is a read-only snapshot, rejects --test, and its --log is console data rather than test XML. --result-file provides a final reader JSON snapshot, which may still say PENDING.

The public run list and raw watcher had a stale membership snapshot. Direct saved run.json, test-results.json, and test-events.jsonl were used instead. Event timestamps are unordered; the helper tracks the maximum timestamp, file modification time, and duplicate first-numbered completions separately. The file-reading API and CLI stdout have size caps, so bounded read-only Python summaries through HardwareControlFabric returned compressed metadata. The helper source is in the evidence branch. The configured certificate filenames here are client_cert.pem, fabric_key.pkcs8, and server_cert.pem; no key contents were printed.

Dependency caches, generated jars, and duplicate console copies are intentionally omitted. A compressed partial console is retained, and the complete terminal reason plus all failed/canceled diagnostics are preserved. The report-challenge tool was not invoked because it automatically enqueues/merges, which this lane prohibited.