TLDR

Resolved as a hard-gate failure by PR #80323's beta.5 correction.

The old body said first-hour-20 had 18 total / 6 pass / 12 fail. That was the pre-correction mock/harness state. The current beta.5 proof shows:

{
  "total": 18,
  "passed": 15,
  "skipped": 3,
  "failed": 0
}

Product impact if OpenClaw moved fully to Codex today: P4 from this issue as filed. The remaining 3 rows are native/live proof gaps, not mock-proven Codex product bugs.

QA impact: P0 resolved for the maintainer mock gate. The gate now has zero hard failures and explicit report-only rows.

Latest Evidence

OpenClaw baseline: v2026.5.10-beta.5
PR: #80323
PR head: 3336dec6419c9cc9a87dc7cfa6f48118ca2d838e
Remote proof run: https://github.com/electricsheephq/openclaw-local-test/actions/runs/25719383976
Confidence tracker: #80936
Artifact: first-hour-20-direct/qa-suite-summary.json

Remote workflow step Run first-hour-20 direct lane completed successfully.

Report-Only Rows

These rows are intentionally not hard-failed in mock mode:

• Instruction followthrough repo contract: mock-openai cannot exercise Codex-native read/write tools.
• Approval turn tool followthrough: mock-openai still models approval followthrough as a Pi-style read call; Codex-native approval/read behavior needs native/live proof.
• Compaction retry after mutating tool: mock-openai cannot create files through Codex-native read/write; compaction replay safety remains a native/live proof lane.

Current Verdict

This issue should stay closed as the stale failing-gate report. Follow remaining proof work in:

• [QA-lab] Codex runtime parity beta.5 confidence proof for PR #80323 #80936 for the beta.5 confidence proof summary.
• [QA-lab] Complete live-frontier token-efficiency and Testbox parity proof #80397 for live-frontier token/native proof.
• Wire optional soak-100 runtime parity lane to scheduled or Testbox proof #80433 for scheduled/Testbox soak-100 proof.
• QA tool-defaults suite conflates Codex-native tools with OpenClaw dynamic tool parity #80319 for searchable/deferred mock-provider fidelity.