refactor(replay): own execution and source data ports in contracts - #4534

Size Report

Startup median (7 runs, lower is better):

This PR is ready at 164b557: I found no problems in the code. CI is green at this head, and the cancelled jobs in the earlier runs were superseded by the passing runs. I did not run tests locally, did not re-measure the startup-median latency left open, and did not read all 60 changed files line by line. Instead I relied on the patches matching the earlier reviewed commit plus spot checks of the contracts exports. Not blocking: the header comments in packages/contracts/src/ad-replay-execution.ts still carry history, and the ../index.ts reference no longer resolves in contracts, so you can trim them to the current ownership statement in a follow-up or leave them. The PR is a draft stacked on t3/maestro-plugin-extraction, so it merges after its base.

The preparation layer remains unchanged at 1078ec04f8baab6301072d20bec94ae0ceb00c55. Its affected admission passed at that exact head (42 passed / 25 skipped, exactCommit=true, checkoutChanged=false) and all 19 hosted checks passed. The retained range-diff shows the preparation patches unchanged through the public-type restack.

I have left the optional comment/constant-placement cleanup out of this round: there is no triggering source defect, and changing this already independently reviewed lower layer would rebase all consumers. The actual contracts-first boundary, moved declarations' runtime consumers, and unused-export/changed-line guard failures were addressed in the published layers; historical failed receipts remain historical failures. This is code/admission evidence only. C3 integration, native ADR acceptance, live/platform requirements and private-package npm setup still block stack integration.

Found 1 test failure on Blacksmith runners:

Failure

^{Need help on this PR? Tag @codesmith-bot with what you need.}

Apex Sentinel · ✅ Approve

Since the last review: 0 new · 0 still open · 0 no longer raised

The changes are the same as on f4a715a, so this is that review, not a new one. Comment @apex-sentinel-staging review to review again.

No open findings.

Reviewed with this repository's own Sentinel configuration (d38e35e61193).

Earlier reviews

The iOS log at 1078ec04 identifies the Blacksmith failure: smoke:deep-link-navigation calls the stored-URL Maestro YAML and receives INVALID_ARGS with details.reason: plugin-not-installed for @agent-device/maestro. The selected Apple aggregate correctly fails with iOS.

A local correction now installs the built Maestro plugin after the shared Apple build, using a job-scoped plugin home inherited by subsequent smoke steps. It belongs in the lower Maestro layer, so it reaches this PR too. Frozen install/build/format and a device-free real installer/source-export check pass; publication is queued in the active stack refresh. Fresh hosted iOS success is still required.

I found no code problems at 8a215d8. Both earlier minor notes are fixed there: AdReplayTargetStructuralDenotation is now Readonly<NodeStructuralDenotation>, and the step-loop.ts comments point to @agent-device/contracts/ad-replay-execution. This follows the earlier review of 1078ec0 (#4534 (comment)).

The two minor threads, #4534 (comment) and #4534 (comment), no longer apply at this head, so you can resolve them.

The checks do not confirm green yet. All 12 jobs that did not pass were cancelled, not failed, which points to superseded runs. A complete CI run on 8a215d8 must finish green before merge. I did not compile or run tests. I judged the type change from the type shapes, and CI owns that check.