Shopify shipped the transactional stack for agent commerce: get_checkout, update_checkout, complete_checkout, with Web Bot Auth identity and UCP consent. An agent can now prove who it is, get permission, and execute the purchase. What the rail does not ship is evidence — whether the product's reviews are real and the store is trustworthy.
// Before complete_checkout, call GoBuy (MCP, keyless):
1. check_store_score({ domain: "any-store.com" })
→ { score: 71, tier: "B", readiness: 68, trust: 76 }
404 = not scored yet → scan_store({ url }) (~20s)
2. check_product_trust({ retailer: "amazon",
product_id: "B0..." })
→ { score: 82, label: "B", state: "indexed" }
→ { state: "not_indexed" } = no score available
3. (optional, deep) analyze_reviews({ retailer, product_id })
→ { review_integrity: 62, adjusted_rating: 4.1,
suspicious_count: 214 } // ~100 reviews, ~1 min
4. proceed / flag / ask the user — then complete_checkoutStore tier A/B (score ≥ 55) → proceed Store tier C/D or unknown → flag to the user before paying Product not indexed → say so; never imply trust suspicious_pct > 15% → surface adjusted_rating, ask user
Rules: quote scores with tier and date. Never present trust scores as purchase guarantees — they are independent evidence for the agent and the human to weigh. See the API docs for full shapes and rate limits.
Identity, consent, execution, evidence — in that order. The first three shipped in September 2026. The fourth is the verification gap, and the agent that fills it wins the trust of the humans it shops for. Full argument: The Verification Gap (position paper) · daily coverage at blog.gobuy.ai.