Finish scalar CI build selection validation

RE-VERIFY — 2026-10-21: Query branch heads and PR state again before continuing. The recorded tests below describe completed local runs, not CI or deployment. The full local suite has no completed result.

Design decisions in force

User requirements dated 2026-10-21: "Keep every existing public contract and test." "No merge/enqueue/deploy." "Never raise timeouts." Run targeted tests locally. Run the complete local suite only through the machine-wide suite-slot.sh wrapper. Open a PR only after the requested test bar; do not watch CI after opening it.

Mission

Finish the deterministic deferred-admission test setup, complete nine-test first-attempt validation and reversal controls, run the complete local suite, and open the why-first PR. The owner requested an underlying correction for unreliable visits to https://droplets.wasmserver.com/. The same host runs kotlin-build-ci. Saved build diagnostics were materialized before narrow selection.

Established facts

The initial six new tests passed 10 independent runs each, 60/60 on their first attempt. Restoring original production source made all six red in three independent runs, 18/18 on attempt 1; all runner repeats also stayed red. Eight nearby existing scenarios passed on their first attempt. These checks preceded the later three-selector extension.

The extension covers shadow identity, startup route validation, and deferred admission. Shadow and route each went red on all three runner attempts, then passed on the first corrected attempt. The first corrected nine-test run had eight first-attempt passes. Deferred admission stopped at its admission-wait assertion in both baseline and corrected runs, so its initial red result is NOT evidence of record materialization. Findings explicitly correct the earlier description. Its setup needs correction before that change can be counted as verified.

The source correction uses derived scalar status fields, early-exit presence checks, point reads for explicit IDs and queued deferred IDs, and matching-row reads for repository cancellation. Independent review confirmed startup cancellation recovery has constant selection passes plus point reads, rather than one selection scan per canceled record. Shadow reconciliation still performs scalar presence scanning per candidate; that is separate from startup cancellation recovery and does not read saved diagnostics.

The complete-suite command was started once through the mandatory slot, then stopped after review found the three additional selectors. No complete-suite XML was produced. No PR exists as of the recorded branch checks. No merge, queue entry, deployment, restart, or artifact publication was performed; the lane used repository operations and local tests only.

Relevant PRs / refs

Next steps

1. Read investigations/laneK1/findings.md and the test-design evidence before changing anything. Re-verify the current branch state.
2. Correct only the deferred test setup so it reaches the queued admission path deterministically. Prove the previous all-history map reads unrelated rows, then verify queued-ID selection passes.
3. Run all nine new tests ten times; require no XML attempt-history. Restore original source plus the behavior-preserving public overloads for shadow/route, then require all nine red in three independent reversal runs. Do not count the old invalid deferred baseline.
4. Rebase, push the exact head, run complete suite via the slot, record first-attempt XML counts, rebase again and run targeted tests before PR creation. Use the exact provenance opening and ending required by the user brief. No merge or deploy. Stop after opening the PR.

Reusable notes

The checkout is under laneK1/kotlin-build-ci. Local validation helpers and XML are under .git/laneK1 and may not survive a filesystem reset; findings and test source are on remote branches. New test names are listed in findings and scripts. baseline-seams.json holds the old-behavior public overloads needed for expanded reversal compilation. The initial six-control branch has original production source and only those six tests.

Repository cloning requires gh repo clone. REST GitHub requests returned a rate-limit response, while GraphQL PR listing and git operations worked. rg was unavailable. Local Git identity is configured. Early checkpoint commands did not create commits because identity was missing; findings correct that, and later remote SHA comparisons verified actual pushes.

Dependency resolution took several minutes when the project source changed. Direct HTTP checks succeeded; do not treat that delay as a test result. Eleven unused dependency declarations were removed from the new deferred test without changing assertions or setup.

Read: both lanePcold7 evidence documents, repository README/AGENTS, Testing Architecture, Engineering Philosophy, and create-handoff instructions. Evidence URLs: https://github.com/CodexCoder21Organization/DigitalOceanDropletWui/blob/wip/lanePcold7-ci-heap/investigations/lanePcold7/findings.md https://github.com/CodexCoder21Organization/DigitalOceanDropletWui/blob/wip/lanePcold7-ci-heap/investigations/lanePcold7/test-design.md https://github.com/CodexCoder21Organization/DocumentationRepository/blob/main/architecture/TESTING.md

Complete-suite command, from the checkout: bash /tmp/claude-501/-code/79f96be0-ae83-4479-b38c-8894e2722ba8/scratchpad/out/suite-slot.sh scripts/test.bash --local --test . --log .git/laneK1/full-suite.xml

Final checkpoint

The last old-map deferred baseline is valid: all three runner attempts reach the historical-row assertion, with each of 129 historical ordinals opened twice. Two earlier fixture runs were not valid baselines; findings explain both. The final corrected nine-test repetition was stopped at the 80-minute checkpoint. Its XML output must not be assumed current; use a new filename for the first resumed run. The queued-ID production correction and final fixture are pushed. The nine-test repetition bar, expanded independent reversal controls, full suite, and PR remain pending. No production action or publication occurred.

K1b pickup — 2026-10-21 22:55 UTC

Read the complete K1 findings and rechecked the handoff. The branch already records the added selectors as corrected; final validation is pending. A fresh clone is in progress under laneK1b. Remaining bar: nine tests ten first-attempt passes each; expanded controls three independent red runs each; final source sweep; exact pushed head complete suite through the shared slot; PR. No production action is planned.

K1b current verification — 2026-10-21 23:04 UTC

The current deferred test is red on all three runner attempts at the historical-row assertion despite the queued-ID map correction. Source review found a separate full-record count/sequence helper used by deferred tracking and marking. A standalone mark test is now included in baseline verification. Further recurring selectors were found in watcher identities, repository counts, retrigger installation lookup, comment rerun head selection, and cancel-all identity/count selection. They are being covered with fail-first tests before correction. These findings expand the remaining bar rather than assuming K1 had already covered them. Product checkpoint: https://github.com/CodexCoder21Organization/kotlin-build-ci/tree/wip/laneK1b-record-selection at 7e978e8cb7ec2843e0a76243d5bf1b89e3627950. This branch has no investigation files in its diff from main. Evidence checkpoint: https://github.com/CodexCoder21Organization/kotlin-build-ci/tree/wip/laneK1b-evidence at d8e4b4b268ae5252649ba5bab09ed633e563b765. Current-source deferred XML is stored there under investigations/laneK1b. No full-suite run or PR has been started in K1b.

K1b validation checkpoint — 2026-10-21 23:34 UTC

Eighteen cases pass ten independent local runs, 180/180 on attempt 1, at beb69ae8c2c5d6fd88004ac0a02a5f3fab654d2e. The final audit then found one standalone sequence-maximum query still reading complete rows; its new public restart test is red at the row-read assertion on all three attempts. Its one-line scalar correction is under targeted validation. Separate test and code reviews report no material issue and a supplemental review confirms the restart fixture and maximum equivalence. Expanded reversal, exact pushed-head complete suite and PR remain pending. Findings and XML continue on wip/laneK1b-evidence. No production action occurred.

23:53 UTC checkpoint

OBSERVED: The complete-suite process is still waiting on the shared slot; its log is empty and no XML exists. Product head 29a9d3c7c09c67f3e87883d84a192f7a1365bdb2 is pushed and the working tree is clean. OBSERVED: Validation completed before the wait: 18 scenarios each passed ten independent runs on attempt 1; the additional restart scenario passed ten independent runs on attempt 1; the final combined run passed 24/24 on attempt 1. The expanded control ran all nineteen scenarios independently three times and produced 57/57 expected row-read assertions. INFER: The remaining dependency is complete-suite capacity. No suite result is claimed. Plan: complete suite, capture XML counts, refresh main and targeted validation, open product PR, complete handoff.

00:12 UTC checkpoint

OBSERVED: The lane has reached approximately eighty minutes. The complete-suite request remains active with no log output or XML result, still waiting for the mandatory shared slot. The brief explicitly permits continuing this wait beyond the eighty-minute checkpoint; the ninety-minute limit remains applicable. OBSERVED: Product source, nineteen scenarios, source sweep, first-attempt counts and reviews are all pushed. The outstanding brief items are complete-suite validation, final latest-main and targeted validation, product PR creation, and handoff completion after that PR opens. INFER: This is incomplete execution pending suite capacity, not a complete validation result.