`verification/dtd-verdict.md`

24 lines
# DTD Verdict — isolate free sample shipping

Decision: A, dedicated sample profile, conditional on a passing pilot.
Confidence: medium. Vote: 2/2 valid; panel availability: 2/6.

| Reference | Runtime/model | Vote | Reason |
| --- | --- | --- | --- |
| Claude | Anthropic / sonnet | Abstain | Disabled by canonical zero-cost mode |
| Codex | OpenAI / gpt-5.3-codex via CLI | A | Sample isolation protects regular freight |
| Qwen | Ollama / qwen3:14b | A | Avoid weight/price heuristics in General |
| Grok | xAI / grok-4.5 | Abstain | Unavailable |
| Kimi | Moonshot / kimi-k2.5 | Abstain | Unavailable |
| Muse | Ollama / muse-spark-1.2 | Abstain | Unavailable |

Dissent: none among valid voters. Four unavailable references limit confidence.

Post-decision Codex debate: KEEP. Prosecutor challenged whether variant isolation
and mixed-cart rate combinations actually work. Defender argued these must be
pilot gates. Judge allowed a bounded pilot only, with expansion conditional on
independent assignment, mixed-cart and large-order evidence.

Outcome: pilot FAILED, so expansion was NOT executed. All pilot assignments
were rolled back. Full outputs: `/tmp/dtd-tk11333-shipping-20260909/`.