[entelecheia]

Refs #345.

The person-activation functional e2e now passes syncDeadlineMs: 60_000 through the existing context seam only for its two expected-success forced cadence calls. 60 seconds exceeds twice the slowest observed completed full sync in the captured hosted macOS job (28.554 seconds). Those child envelopes are 90 seconds, leaving 30 seconds for startup/stop/cleanup; all other child envelopes remain 60 seconds. The named test envelope is 240 seconds for two cadence budgets plus activation, plain sync and cleanup. The diagnostic asserts both overridden calls actually observed 60 seconds.

Every existing writer-drift, floor, backoff, remote-row and exact-marker assertion remains. There is no retry or relaxed result assertion, no product source change, no CLI flag, and no change to the product's 30-second default or stop/retention/fail-closed behavior.

The captured failure at head 0a6bcc2, hosted macOS job129793329365, established call12's own deadline, not a cumulative budget from earlier calls:

Call12 phase	Child-relative seconds	Span seconds
stop created, 30-second deadline	0.241
transport entered	0.546
before plan preparation	9.697	9.150 after transport entry
plan fsynced	17.664	7.967 including guards and durable preparation
artifact built	24.672	7.008
before CAS	28.715	4.043
deadline fired	30.243	30.002 after stop creation
CLI finished	30.523

At before-CAS, 94.912% of the budget had elapsed and only 1.526 seconds nominally remained. The marker precedes installed-object/final guards and the actual push, and no CAS-returned/ack/import/receipt marker was observed. Therefore the actual remote effect is unmeasured; the returned ok:false, outcome:unknown and retained attempt are the correct conservative response. Earlier calls10 and11 took 5.033 and10.314 seconds independently and did not consume call12's timer. These phase spans do not isolate a syscall, device or scheduler cause.

A focused real-deadline fixture regression aborts after durable publication intent, proves false/unknown with an unchanged retained plan and intent, no completed receipt or remote refs, and then resumes that same plan successfully. Existing sync-cadence deadline/group/backoff tests remain unchanged.

Product-level follow-up observation only: a full placement sync can take about 30 seconds on a slow hosted runner (plain call3 completed in28.554 seconds). No product action on that observation is included here.

Validation status: implementation head 7b0f87f is supplied as a draft for hosted CI. No product source diff and diff-check is clean. The initial remote typecheck at b6f63eb failed only on a shared/non-shared Buffer expectation type; byte comparisons were changed to Buffer.equals() without changing their meaning. Final-head typecheck, boundary, focused/full tests (including the new deadline regression) and pack are unverified: retry admission was deferred with three, then four other verify jobs against the cap of three. No extra run or heavy Mac fallback was started. Hosted CI must validate this head before acceptance. The issue stays open until 10 hosted macOS passes after the fix are recorded; that evidence is not claimed by this PR.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has insufficient remaining quota. Try requesting a review at a lower effort level.

[chatgpt-codex-connector]

Codex Review Summary: ✅ Completed

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-11-10T15:45:46.511242Z	7b0f87f	Draft marked ready

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.