Published result
Approval delays lack phase attribution
A plain-language summary comes first, followed by the full text of the result's reproduction pack.
Result summary and pack text · reproduce/approval-delay-evidence/README.md
- Finding id:
approval-delay-evidence - Verdict: Unresolved
- Finished: 8 November 2026
- Comparison: Test setup, rules and sandboxing
- Reproduction pack: reproduce/approval-delay-evidence/ (data, analysis script and rerun steps)
- Category page: categories/harness-levers.md
In plain terms
Two approvals occupied their Build slots for more than 94 minutes. The available records do not identify the blocking call; individual readiness commands already had time limits.
Verdict key: Yes, The result supports the change or claim that was tested. No, The result does not support it, so nothing changes. Unresolved, The evidence is not strong enough either way. Superseded, A later test replaced this result.
Full text of the reproduction pack
The text below is the pack's README as published in the repository (a system folder name may show without its leading slash, so that the site's privacy scan passes).
Finding. Two approvals took 94 minutes 18 seconds and 97 minutes 54 seconds from the wrapper's approval marker to its Build marker. Both eventually reached Build. The retained evidence does not identify the blocking call or establish that either process was deadlocked.
Evidence
The Studio wrapper marked approval at 05:10:51 UTC and Build at 06:45:09 UTC. The MacBook marked approval at 05:04:51 and Build at 06:42:45. Both logged Kogen v0.1.0-dev.62015185. Just before approval, Shape reported readiness times of 68.1 and 22.4 seconds respectively. The wrapper ran two approval invocations: it captured the first card with a pipeline, then printed only the last line of the hash approval. It did not timestamp their boundaries or stream the first card to the build log.
The Studio first-card output file's last modification was 05:57:00 UTC. A retained Studio approval readiness attempt started at 05:56:10 UTC, took 49.521 seconds, and recorded snapshot 0.130s, materialize 1.160s, setup 5.744s and 0.038s, feasibility 1.907s, model checks 1.286s in total, acceptance 37.354s and step analysis 0.384s. Acceptance's command status was failed while the attempt outcome was passed; this is a readiness dry run, and its status alone does not diagnose the delay. This record covers only that attempt, not the whole wrapper interval. It leaves about 45 minutes before the recorded readiness attempt and about 48 minutes between card-file completion and the Build marker without phase attribution. The MacBook card output file was last modified 06:03:20 UTC, leaving roughly 58 minutes before that point and 39 minutes afterwards. File modification time bounds the last write, not the process exit or the exact start of the second invocation. No original process tree or stack was available at inspection.
Source and limits
The Kogen release pins shaping 7bcfc8d32b83 and verification 67617548616c. Approval opens the repository, acquires its build lock, and then performs pre-readiness work; an uncached approval calls readiness Check. Readiness runs setup, feasibility, model and acceptance checks. The session context is cancellable but the whole approval has no deadline in the shown path. Contrary to an assumption that all commands are unbounded, individual readiness commands have a 120-second timeout. Observe phases have no bound; other pre- and post-readiness calls also need inspection. None is proven to have caused these incidents.
The wrapper has no overall timeout. A bounded approval with context cancellation, process-group cleanup and a specific refusal reporting invocation, active phase and elapsed time would limit future stalls; preserve the existing readiness gates. To establish the actual cause, retain timestamped start and finish events for both invocations and each pre-readiness, readiness and publication phase, plus a non-signalling process tree and wait-channel or stack evidence while the delay is live. Capture child exit and cleanup status. The retained attempt only records phases after it begins and cannot explain earlier time. No fault reproduction was performed because the blocked dependency is not identified; fabricating one would demonstrate a possible failure, not this incident's cause.
Rerun
Run python3 report.py to recompute the recorded wrapper intervals and print the evidence gaps from the included data. The original production logs and private attempt are not distributed; the selected times and durations are transcribed above. Inspect the linked pinned source for call paths and timeouts.