[AlexanderZ-Band]

Scope

P1 of INT-1850: synchronous snapshot installation and exact subscription/attempt cleanup in the Phoenix client. Optional reconnect participation follows in P2 #83. No platform, SDK/Core or REST changes; no package publication.

What

Add/export on_join_response, invoked synchronously before readiness and deltas across initial, socket and channel joins. Match attempt refs on both protocols, retain V2 filtering, classify lifecycle failures and fail closed on pre-install queue loss. Preserve caller-owned initial admission and cancellation-safe captured cleanup awaited by shutdown.

Validated at 3d39524: full pytest --cov 442 passed, 97.60% coverage; focused gate, locked sync, Ruff, format, Pyrefly and every pre-commit hook passed (all exit 0). Installed-wheel real-wire smoke passed initial/reconnect snapshot, readiness, deltas and cleanup; wheel sources match this head. Real-wire/controlled-synchronization regressions cover the finalized plan scenarios and compound flows. Band review converged; all feedback fixed and resolved. Author system/ownership/simplicity review complete. Full local suites also passed on Python 3.12 and 3.13. The backoff timer observer was verified and corrected at the native call_at boundary for supported Python versions. All required current-head CI passed: Linux Python 3.11/3.12/3.13, Windows Python 3.13, lint, types and PR conventions. Native stack #84 submitted to the main merge queue.

Why

Join responses must install the consumer snapshot before events can use it. Exact attempt settlement and cleanup prevent stale replies, cancellation or replacement from corrupting admission.

[linear-code]

INT-1850

[AlexanderZ-Band]

Blocking — would be REQUEST_CHANGES on a PR by someone else.

Six-angle Band review of P1 at fe70262c196cf1ac2f51457717d5d4db81301acd (against main). Goal and owning layer are supported; initial triage is DRAFT.

Two functional defects were reproduced through public APIs and the real wire peer on both protocols: terminal close bypasses retained cleanup during shutdown/context exit; established installer failures lose their exception cause on both recovery paths. All six reproduction cases completed with resources released.

Inline findings: A 2 nits; C 1 issue; D 6 issues; E 1 issue; F 1 issue. Existing focused hook/lifecycle suites pass 72 tests, which do not cover these failures. Review-only: no code changes. Supported-platform CI remains skipped for the draft. Optional recovery is reviewed separately in #83.

[AlexanderZ-Band]

Blocking — would be REQUEST_CHANGES on a PR by someone else.

Cycle 2 full six-angle review at f1c15b479003b19b3e40b960c24a2bd292d1483a: production goal/ownership remains supported and no new runtime flow/logical defect was confirmed. Remaining findings concern causal test proof and small reuse nits; fixes are authorized and will be verified before convergence. Both PRs remain drafts.

[AlexanderZ-Band]

Blocking — would be REQUEST_CHANGES on a PR by someone else.

Full six-angle review otherwise found no actionable findings at this head. The independent skeptic reproduced this send/reply settlement defect on both wire protocols, verified the dependency flow-control event source, and confirmed cleanup. The goal and existing attempt-future owner remain supported; a localized first-outcome arbitration correction is required before merge.

[AlexanderZ-Band]

Blocking — would be REQUEST_CHANGES on a PR by someone else.

The fresh full six-angle review found no further production defect. This remaining explicit resource-contract test gap is being corrected and verified; one bounded additional review round will establish convergence rather than merge on an unreviewed final fix.