The merchant sees the agent's network identity (datacenter vs residential IP, ASN), transport fingerprint (TLS/HTTP version), header presentation (user-agent coherence, sec-fetch-* headers, Accept-Language), funnel behaviour (did it view the catalog before checkout, pacing between steps), any Web Bot Auth signature, and — on payment — wallet history, settlement latency and amount exactness. These are scored to classify the buyer as browser, honest automation, or hostile script.
Most common causes: a Mozilla user-agent with no sec-fetch-* headers (impersonation mismatch), tool-default user-agents like python-requests or node-fetch, datacenter IPs, millisecond pacing between steps, and skipping the catalog. Each is an automation signal default bot filters act on before checkout completes.
Run the agent through a real instrumented checkout that returns the merchant-side forensic file instead of a product. Glass Merchant does exactly this: a free presentation-only run returns your header and transport forensics; a paid 1 USDC run on Base adds payment forensics. Real merchants never open their fraud dashboards to you, so an instrumented store is the only mirror.
Glass Merchant — run your own agent through an instrumented checkout →Operated by Last Minute Deals HQ LLC. This page is a plain-language answer; the linked service is live and machine-payable.