id: hf-2026-09-20-finish-and-verify-the-runner-shared-preparation-ledger url: url://handoff/handoffs/hf-2026-09-20-finish-and-verify-the-runner-shared-preparation-ledger title: Reconcile shared-preparation ledger findings after PR 118 owner and remote allocation work summary: The preserved ledger branch overlaps open draft PR 118, which remains with the active parent workstream; remote test allocation is also an open prerequisite. The upload-capacity dependency is complete and removed. Revisit after those two dependencies resolve. created: 2026-09-20T10:30:32.213Z completed: null blocked-reason: Assigned brief requires landed cancellation design or an orchestrator ruling lifting the sequence; prerequisite https://github.com/CodexCoder21Organization/BuildTestRunner/pull/122 is still open. dependencies:

url://handoff/handoffs/hf-2026-09-20-restore-remote-test-allocation-for-the-runner-preparation-checks
url://handoff/handoffs/hf-2026-08-24-fix-startup-compaction-delta-heap-exhaustion-then-revalidate-the-extreme-suite

Reconcile the shared-preparation ledger against BuildTestRunner PR 118

Verified 2026-09-26 by sweep4-T1. Keep the ledger branch as a reference while the live parent workstream owns overlapping PR 118; the validation-allocation dependency remains open. The completed upload-capacity link has been removed.

Remaining work, in order

Wait for the active parent workstream to finish its current PR 118 decision: either absorb the ledger findings into the runner work or land PR 118 before rebuilding any findings. The parent handoff has a live owner and its current body still lists PR 118 validation and the pipeline transition.
Wait for the remote test-allocation handoff to complete the current allocation work before attempting any required remote runner validation. Its latest checkpoint records a 30-second listDroplets failure before tests; do not treat it as evidence against the ledger branch.
After those dependencies resolve, re-evaluate the preserved ledger branch against current main and the final PR 118 design. Its earlier review found an unimplemented close-task shutdown race, known one-lane parity regression, and missing invariant coverage. Preserve the unchanged one-lane contract; do not reuse the old candidate as verified work.

Verified current state

BuildTestRunner PR 118 is OPEN and draft at eed4402c91081ec55a0fa64311c2e748e70289a3, branch wip/LPIPE-d-pipeline-handoff, with merge state BLOCKED and both checks FAILURE. GitHub last updated it on 2026-09-24, outside the two-hour window.
The preserved reference branch fix/pipeline-one-preparation-ledger remains at 934807339edea0f858768fa034ad54683a1d183e; it has no PR and has not changed since the prior snapshot. It overlaps PR 118's lane-ticket registration files and findings, but is not itself ready to land.
The active owner handoff is the parent startup-compaction and utilization workstream. Its live body still lists PR 118 validation; claims show fable-code-2026-09-21 live at 2026-09-26T23:09:58.589Z.
The open validation prerequisite is remote test allocation for runner-preparation checks. Its current body is being worked by W11; it describes the open BuildTestEmbedded PR 1097 and a current candidate branch wip/provisioning-deadline-W11 at a9a08953ca09e5002c322c94cc0096b7c556c447. PR 1097 remains OPEN/DIRTY at older head fb3c005344613ac74415bf1b3485342c078ae3b5; no change was made to that work.
Remote upload-capacity handoff is completed as obsolete on 2026-09-24; its body says uploads had resumed and points to the separate allocation handoff. This stale dependency has been removed.
The PR 118 required run 171cd8cb currently exposes 255 PENDING results with reconciliationPending=true; that endpoint does not give a final per-test verdict. No run was requested by this lane.

OBSERVED / INFERRED

OBSERVED: target claims were empty before this lane claimed it. PR 118 and its reference branch match the states above. The parent has a live non-sweep4 owner; the upload-capacity dependency is complete; the allocation handoff is still open and has an active current checkpoint.

INFERRED: the parent handoff remains the right owner dependency, and allocation remains a validation prerequisite. The completed upload-capacity dependency was stale. No code, test run, production action, merge, or enqueue was performed.

Update 2026-10-01 (board-drain orchestrator)

The remote test-allocation handoff (hf-2026-09-20-restore-remote-test-allocation-for-the-runner-preparation-checks) is COMPLETED: BuildTestRunner remote runs allocate and execute again (runs 19221bf5 270/270 and 9628a572 277/277; no listDroplets stall in the newest 500 runs). Item 2 above no longer waits on anything.

Verified firsthand 2026-10-01 01:50 UTC: the ledger branch's test testEvictedWarmAdmissionStaysClosed still fails on BuildTestRunner main b2109d0e with WarmChildAdmissionClosedException: Warm child admission for 'second' in 'second.kts' was closed before launch. Probe branch (main + that one test, diagnosis only): https://github.com/CodexCoder21Organization/BuildTestRunner/tree/wip/K56-evicted-admission-probe at afb14ba7c6bba28c27096b097675ae3727b38f64. Mechanism: the warm-child admission callback captures the mutable PipelineGroup (PreparedLeasePipeline.kt); closeWarmChildAdmission marks the group closed and warm() resets the flag when the group is prepared again (~line 641), so a callback left over from an evicted round closes the next round. The closed state must belong to one preparation round, not to the group. Sequence the fix after https://github.com/CodexCoder21Organization/BuildTestRunner/pull/122 (which rewrites the same code across ~40 files); port only the per-round-state part of the ledger branch's fix commit 2657640, since that commit also changes the expectations of testLateWarmFailureAfterCancellationIsPreserved, which needs a separate design decision.