`verification/dtd-discovery/codex-debate.txt`

9 lines

**PROSECUTOR:** A risks mistaking a workaround for a verified launch path. Native collection visibility does not establish that customers can discover Newmor through ordinary search while Boost returns nothing. An exact-name fallback may serve only shoppers who already know the vendor. Two passing canaries cannot establish that all 116 products have correct visibility, concealed material prices, and working purchase or inquiry paths. “Live Shopify data” establishes provenance, not correctness. Only two models voted; the missing four provide no supporting evidence. A could therefore authorize a launch with inadequate coverage disguised as verification.

**DEFENDER:** That objection defeats unconditional publication, which A does not authorize. A explicitly limits publication to verified products, makes 116 a ceiling, and requires browse and transaction-path checks before rollout. Canaries test the integration; they cannot substitute for verifying each released product. Exact Newmor search supplies a bounded discovery route without changing other vendors, and the native collection addresses the immediate browse failure. Broader search remains imperfect, but the supplied constraints do not require complete Boost recovery before any release. Waiting supplies no demonstrated remedy, while publishing everything abandons the verification constraint.

**JUDGE:** The prosecution identifies a real distinction between testing two canaries and verifying the release set. However, A’s explicit “only verified products” condition resolves that distinction: passing canaries alone cannot authorize the remaining products. The evidence supports choosing this conditional approach, not declaring its implementation ready. The unavailable voters reduce confidence but do not invalidate the argument.

FINAL: KEEP

A provides a bounded route to launch while preserving the stated constraints: verified products only, no more than 116, Online Store only, concealed material prices, functioning customer paths, and no new spend. The strongest objection concerns insufficient verification, but A already makes verification a prerequisite; it does not authorize treating two successful canaries as proof for the entire catalog.