test1203 /btw production E2E gate
Date: 2026-08-26
Base: PR #1202 head 8ad03b1ea3c9fcfae085d85070213c9d2dffb5ac

Scope:
- real Codex 0.148 owned-stdio active-main / double-fork / reverse-completion / cancel isolation: delegated to reviewed test1190 live gate
- durable command ACK loss and restart: executable offline gate
- terminal WAL response loss and restart-before-pull ordering: executable offline gate
- bring-back accepted replay exactly once and ambiguous response fail-closed: executable offline gate
- attachment node-token / byte-count / SHA-256 / 0600 materialization: executable offline gate
- persisted receipt redaction: executable offline gate

Honest gap:
The production Codex adapter does not yet serialize materialized attachments
into turn/start. Therefore this gate does not claim model-side attachment
consumption. It is a merge gate for the follow-up production wiring, not proof
that wiring already exists.

Executed evidence:
- Docker image `anet-test1203`, pinned `codex-cli 0.148.0`
- registered L1 image is bound to the exact 40-character lowercase source SHA;
  missing and malformed bindings are both witnessed-red
- offline recovery layer: 5 passed, 0 failed, 20 assertions
- live owned-stdio layer: PASS, reviewed evidence `test1190-wire-v2`
- source active at fork; three threads concurrently active
- cancelled target `interrupted`; source and sibling `completed`
- created slow then fast; completed fast then slow
- witnessed-red source-first mutation rejected
- disposable CODEX_HOME files after exit: 0
- sanitized wire trace scan: no prompt markers, local home paths, tokens,
  bearer values, or raw UUIDs
