[RCmerci]

Startup published the LUI root immediately after native admission, allowing SwiftUI initialization to delay local account restoration. Buffer admission-time callbacks, deliver them after native start/connect, and publish the root after the initial platform tasks finish. Generation checks fence cancellation and stale responses; the host retains the starting runtime until cleanup completes.

Validation on 6684ce28e094753d1579511fc230d47581f75862:

• New startup boundary tests: 10 scenarios, including owned C bytes, worker callbacks, repeated start, stop/cancel, stale cancellation, rejected admission and nil responses; the pre-change baseline fails the root-publication assertion. Existing runtime failure tests: 11 scenarios. Both runners are now in CI.
• Full local dune build @all app/native_embed.exe.o, dune runtest --force, and Swift platform/account recovery tests passed. Independent exact-head review found no blocking new defect.
• Fresh process restarts, same synthetic fixture, alternating order, equal untimed prewarm after each install: 10 pairs. From production prepare.begin to the first actual row onAppear, medians 519.255 → 380.803 ms (10/10 pairs faster); Loading visibility proxy 266.525 → 191.478 ms. A separate 10-pair install-transition batch is retained; results are not pooled. Screenshots/profiler are excluded from timing.
• Real native cancellation, startup-time leave/return, absent account and binding-error/restart recovery passed. UIKit Timeline → detail → back passed (controller/path counts 1/0 → 2/1 → 1/0).

Limits: Simulator iPhone 13/iOS 26.1, Swift Debug/OCaml dev, experimental-host-restamp, cross_abi_validated=false. These are lifecycle proxies, not GPU presentation or device Release results; the paired iPhone is unavailable.

Known baseline boundary: immediate synchronous stop/start can reject the second native admission (reproduced with original main Runtime, 1/3 attempts). Asynchronous worker disposal is a supported explanation but its internal state was not captured. The final candidate stop comparison passed 3/3, while two earlier unlocalized stop failures remain recorded. This PR does not claim to fix that existing native restart race. Five pre-existing agent-document format errors also remain unchanged.

[chatgpt-codex-connector]

Codex Review Summary

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

Review	Status	Commit	Review trigger
📝 Code Review	🔄 Running since 2026-10-22T03:03:29.656958Z	6684ce2	PR opened

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.