Core required cases pass 210/210; full suite has five attempt histories. Provider-move fixture cause is isolated; four intermittent results still need reliable reproducers.
Claimed by laneB16
RE-VERIFY: Current source is https://github.com/CodexCoder21Organization/UrlResolver/tree/wip/laneB16-finish-1336 at ef3cb5fb3323a015740a1cbe9481bd6f85644c8c. PR https://github.com/CodexCoder21Organization/UrlResolver/pull/1336 is still open at its previous head; it has not been updated yet.
Original provider-move case has now reproduced on attempts 1 and 2 at line 206, before passing on attempt 3. Original closed=true, replacement settled=false, active connection=true; separate outbound endpoint appears in cleanup. Its XML is https://github.com/CodexCoder21Organization/UrlResolver/blob/wip/laneB16-original-evidence/.laneB16/provider-original.xml. The fixture-only commit adds the neighbouring registration gate while preserving all 317 original lines, assertions and bounds. First three corrected invocations are clean; the ten-run gate continues. Next: finish ten runs, 20 interleaved candidate/main pairs for the registry-address setup case, focused set five times plus the changed fleet test, one full local suite, classification, final PR head/body update. No production operation or publication.
Earlier snapshots retained below; this section is the current B16 state.
RE-VERIFY: 2026-10-21 00:35 UTC. The original seven neighbour results and 17 required-check cases are corrected, but the complete-suite gate is not clean. The PR remains open/unmerged at 89542f8ddde83a38645df695ea95af781dce30c5; no branch/body update was made. Resume from source 980c7bc547dcd544e59aaa9e4f2531a43f6d264a at https://github.com/CodexCoder21Organization/UrlResolver/tree/wip/laneB15-rebased-source . The working tree is clean. No publication, production update or restart, check request, merge or queue action was performed.
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.
Finish five full-suite cases before updating https://github.com/CodexCoder21Organization/UrlResolver/pull/1336 . Preserve every assertion/count/bound and use local targeted tests; no production action or publication. Core correction mechanisms are established: mixed control/RPC offers wrongly enabled application setup classification; diagnostic probes wrongly marked pooled roots; missing local handlers were relabelled as physical setup errors; ordinary address errors lost their original wait wrappers. The source corrects those distinctions while preserving genuine RPC setup marking and pending-handshake timeout classification.
The complete suite on 39071580a3c59dd46468e89cf58d33eff3638ca8 ran 2202 cases:2197 first-attempt successes,2201 final successes,five histories containing 11 attempts. XML full.xml and full-results.json in the evidence below contain complete details. Focused gate on that source is210/210 first attempts across 42 cases five times; six controls red on all18 attempts; corrected union18/18. No previous focused/full XML may be relabelled as proof on a newer head.
A required pre-build main rebase moved the candidate to980c7bc547dcd544e59aaa9e4f2531a43f6d264a. New main is dae99386f9e3aefedffc3020242d25547ecda994; its sole change since018f3b07ce1f77bf21fa3470ded79fb4df5d23c7 is tests/stressTestRelayRegistrationRetriesSlowBootstrapBeforeLowerTiers.kts. Production source, build definitions and artifact declarations are unchanged. The five follow-up cases are byte-identical to main. Candidate follow-ups and main comparisons are separate XMLs; see suite-comparison-audit.json and main-comparison-head.json.
testActiveQueryRetainsConfiguredAddressAfterProviderMoves: full run red 3/3 attempts; candidate follow-ups red 5/5 first attempts then later pass; main red 3/5 first attempts. Its close wait at line 206 sees a NEW outbound connection to the MOVED endpoint, after the original connection closed; the watched replacement-admission hook is not reached. The fixture cancels periodic exchange but independent relay registration can still open this connection. A neighbouring two-address fixture already gates that registration work. Neutral scenario testProviderMoveKeepsAddressSelectionVisible copies the original assertions and bounds, adds only public connection observations, then holds the existing registration gate. The gate is observed, and the held version passes 5/5 first attempts with zero histories. Red source https://github.com/CodexCoder21Organization/UrlResolver/tree/wip/laneB15-provider-observation at 2ec0cc1dcdac68c36e935212c153b0e4bad752b1; held source https://github.com/CodexCoder21Organization/UrlResolver/tree/wip/laneB15-provider-held at 9bf9e44d4d55cf2ef48dfa17fd7048f12f686186. XML provider-observation-1.xml (first red, later pass), provider-held-1.xml and provider-held-more-1..4.xml (five first-attempt passes). Original existing fixture is NOT edited. Next: review and apply the same fixture precondition while retaining all assertions; record its reversal.
Do not call these resolved or pre-existing solely from those green targeted samples. After fixing or positively refuting each, rerun the42-case focused gate and one complete local suite from the exact source proposed for the PR. Only with zero histories: fresh PR-open check, fetch/rebase, lease-protected push to wip/laneB2-reconnect-contract, then gh api PATCH body.json. Existing make-body.py/update-pr.py are guarded success-path helpers; do not use old summaries as a new source proof. No CI observation after the PR update.
All current remote heads were queried during this write. Source checkpoints are explicitly separate from diagnostic scenarios and evidence-only branches. Nothing is published or deployed; all code is unmerged.
Final combined evidence is being preserved at https://github.com/CodexCoder21Organization/UrlResolver/tree/wip/laneB15-final-evidence/.laneB15 . The local main-comparison branch has no unique source edits and matches main dae99386f9e3aefedffc3020242d25547ecda994. Exact process audit reports zero remaining lane test/CLI tasks; raw process arguments stay local.
The earlier B14 record is retained below as historical input; its remaining-seven section is resolved by B15's core correction, but the broader full-suite gate above remains unmet.
RE-VERIFY: This is a snapshot from 2026-10-20 around 23:15 UTC. Recheck the PR state/head and remote branch heads before editing. Do not assume this snapshot is current. Read the linked findings and complete XML records before running anything. Do not reuse this checkout.
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 combined two already-developed changes for https://github.com/CodexCoder21Organization/UrlResolver/pull/1336. The broad gate returned seven errors, so the requested PR update was not performed.
Explain and correct the seven candidate-only neighbour results, prove each with an initially red public-API scenario, review the test coverage and implementation, then satisfy the requested gates. The user requested the neighbour group once and full suite once from the exact source head pushed; the first neighbour invocation is already recorded and did not meet the bar. Agree the next gate schedule with the manager after the seven mechanisms are understood. Do not describe a rerun as the original invocation. Do not run the full suite while iterating. No test limits or iteration counts may be relaxed.
The seven exact selectors and complete errors are in baseline-seven.json, neighbour-results.json and neighbours-1.xml in the evidence branch below. The results were: queued recovery cancellation began replacement work before the queued runnable; record replacement emitted a setup warning; handshake diagnostic wording differed; synchronous stream setup returned a different exception type; completed recovery admitted an inbound stream; queued recovery began connection setup before dispatch; relay completion produced an unexpected TCP forwarding error. Each returned an error on all three runner attempts. No shared mechanism is established. All seven test files are byte-identical to main. Main comparison groups 1–5 each passed all seven scenarios on the first attempt, 35/35 with no histories, at main 018f3b07ce1f77bf21fa3470ded79fb4df5d23c7. Those results do not support a pre-existing-main explanation. Complete XMLs are in the baseline evidence branch.
B13 input was 89542f8ddde83a38645df695ea95af781dce30c5. B12 input was d4f0dad4a72de928f7d0ed714f2cb3eb5901fbce, based on 0505f67250def3c9aa04cef85fbb0d5ea24d650f. Four B12 commits were replayed onto B13. There was one textual conflict in src/foundation/url/resolver/RetirementAwareRelayHost.kt, at the openRecoveryStream argument list (original marker lines 277–283). The result preserves callerStopped from B13 and initial liveParentExists from B12 in signature order. UrlResolver.kt merged without a textual conflict. B13 admission owns the setup-error transition and caller-stop semantics; B12 unfinished-handshake expiry remains TimeoutException with the exact diagnostic. Recovery expiry uses the initial recovery phase context; ordinary admission retains B13 phase transitions. Read both input findings in the final evidence branch for their original proofs.
Current candidate is 262f23d5f1bdd989b4657fe84e02a17a5c0c0285. Production source is byte-identical to merged production commit 2463dab2647bd2a969a55c4370cda439d7ea49ef. The later change adds only 18 lines to tests/testSandboxedObjectReleasesResourcesAfterReconnect.kts, retaining all 197 original lines in order. The original fixture could observe setup on a different physical parent and then wait for its original healthy parent to close. The updated fixture selects the exact original connection through the existing public bridge, performs the child setup on that connection, and observes the actual proxy worker before continuing. No production tracing remains. A temporary 100-cycle diagnostic returned the original close wait on all three attempts and showed the original parent had never received the setup classification. That diagnostic is archived separately and is not a discovered test on the candidate.
Final focused gate on the exact candidate: 29 unique scenarios, five invocations each, 145/145 first-attempt successes, zero attempt histories. XMLs review-focused-1.xml through review-focused-5.xml and focused.json are in reviewed-focused evidence. Reversal controls: six cases, all six first attempts red, all 18/18 runner attempts red; controls were on identical production source and then restored. An independent review found no remaining concrete integration or fixture issue after the explicit proxy worker observation was added. This review does not establish a mechanism for the seven later neighbour results.
The neighbour selector preserves B11's 449 core scenarios plus three preservation scenarios, 452 unique actual function names. One invocation: 445 successes, seven errors, 21 error attempts. Manifest SHA256 90dcc103bbc75bff994df37f69a1762d75761fd1223a33e691c570d935971675. Full suite was not started. The PR branch/body were not updated. Last REST read: PR open at 89542f8ddde83a38645df695ea95af781dce30c5. Main was 018f3b07ce1f77bf21fa3470ded79fb4df5d23c7; repeated main rebases left the source candidate unchanged.
Evidence files are under .laneB14/ on evidence branches. Local working files were under .git/b14/, excluded from scenario discovery. Download evidence rather than placing another active checkout inside the checkout that runs tests: kompile also discovers nested source under .git, which caused one setup-only invocation to execute zero scenario bodies. That setup invocation is excluded from positive counts and recorded in findings.
Clone fresh under your own lane directory with gh repo clone. Read README and AGENTS if present, the input findings, and https://github.com/CodexCoder21Organization/DocumentationRepository/blob/main/architecture/TESTING.md plus https://github.com/CodexCoder21Organization/DocumentationRepository/blob/main/PHILOSOPHY.md before test changes. Read all seven complete XML records without reproducing raw log output in user messages. The invariant table must precede ownership/lifecycle/concurrency changes. Keep the public API unchanged unless separately established. All local scenario commands use scripts/test.bash --local --test <selector> --log <xml>; inspect attempt histories, not just final success. Only one full suite may run on the machine at a time. Check actual process arguments without matching your own shell text; bounded waits only. Fetch and rebase main before builds and before final push.
Once the requested gates are satisfied, check that the PR is still open, fetch/rebase main, push the candidate to wip/laneB2-reconnect-contract with lease protection, then update the body through gh api --method PATCH repos/CodexCoder21Organization/UrlResolver/pulls/1336 --input body.json. Keep the exact owner provenance, B12's proposed classification section edited for accuracy, and an accurate Round 6 paragraph. A saved body draft has a placeholder and must not be submitted unchanged. Do not use gh pr edit. Do not watch CI afterward, request checks, merge, enqueue, publish artifacts, deploy, or restart production services. Nothing from this lane was deployed or published. All source is unmerged.
The lane stopped at its explicit time bound with durable source and evidence. This is remaining engineering work, not a credential or service dependency.
79f96be0
2026-10-21T00:59:42.818815+00:00 RUNNING — finish the provider fixture, compare the registry-address case, validate and update https://github.com/CodexCoder21Organization/UrlResolver/pull/1336. The PR is open, unmerged=True, at 89542f8ddde83a38645df695ea95af781dce30c5. Source checkpoint ef3cb5fb3323a015740a1cbe9481bd6f85644c8c is on https://github.com/CodexCoder21Organization/UrlResolver/tree/wip/laneB16-finish-1336; it is not yet on the PR branch. The fixture now holds independent relay registration during the provider move. All 317 original lines remain unchanged. The corrected scenario passes 10/10 first attempts; original red evidence and corrected XML are remote. Comparison progress: {"UrlResolverCandidate": {"runs": 9, "first": 6}, "UrlResolverMain": {"runs": 8, "first": 8}}. The 40-sample comparison continues; separate neutral debug observations are excluded. No product change in B16. Delegated work: none. Holds audit: final gate waits for the requested rate decision; no user answer needed. Landing sweep: no merge or queue action permitted. Owner direction captured: existing brief requirements remain recorded in findings and handoff. Next steps: finish the rate comparison and any proven source correction, validate the focused and complete suites, update PR head/body and handoff, then stop all lane processes. Lessons: count XML attempt histories by test name; isolate independent fixture work with an observed boundary; verify main-rate claims against actual XML before classifying suite histories.
79f96be0
79f96be0
79f96be0
79f96be0
Moves it out of every priority list and into ArchiveArea.