[aw-failures] [aw-fix] Daily Cache Strategy Analyzer (main) 100% red — experiment picker feeds Codex an invalid model name (isIn [Content truncated due to length]

[github-actions]

Fix the experiment model picker in daily-cache-strategy-analyzer — it feeds an unsupported model name to the Codex engine, hard-failing 100% of runs it selects.

The workflow sets engine.model: "${{ needs.activation.outputs.model_size }}", populated from a pick-experiment step (GH_AW_EXPERIMENTS_MODEL_SIZE). When the picker selects a model Codex does not recognize, the harness aborts immediately with isInvalidModelError=true and refuses to retry, so the whole scheduled run fails.

Affected workflow / runs

• Daily Cache Strategy Analyzer (.github/workflows/daily-cache-strategy-analyzer.lock.yml, engine codex v0.142.5, branch main)
• Representative failed run: §28967575350

Evidence

[codex-harness] attempt 1 failed: exitCode=1 isInvalidModelError=true ... retriesRemaining=3
[codex-harness] attempt 1: invalid/unsupported model configuration — not retrying (specify a valid engine model name in workflow frontmatter)
[codex-harness] done: exitCode=1 totalDuration=7s

Probable root cause

The experiment model pool (pick-experiment → model_size) contains at least one value that is not a valid Codex-compatible model id for engine version 0.142.5. The picker output is passed straight into engine.model with no validation, so any bad draw is a guaranteed hard failure.

Proposed remediation

• Constrain the experiment model_size candidate list to model ids known-valid for the pinned Codex engine version.
• Validate the picked model against the engine's supported set before launch; on mismatch, skip the experiment (noop) instead of hard-failing the run.
• Optionally, treat isInvalidModelError as a skip/soft-fail for experiment workflows rather than a run failure.

Success criteria

• 5 consecutive scheduled runs complete without isInvalidModelError.
• An invalid experiment draw results in a skipped experiment (green run), not a red failure.

Related sandbox-firewall failures this window are tracked separately in #44242.

Analyzed run: 28967575350.

Generated by 🔍 [aw] Failure Investigator (6h) · 211.3 AIC · ⌖ 34.6 AIC · ⊞ 5.2K · ◷

• expires on Jul 15, 2026, 11:25 AM UTC-08:00