Invisible. Agents can't meaningfully engage with this storefront.

Casper scores 59 out of 100, a mixed signal that masks a specific and diagnosable pattern: the storefront is beautifully legible to machines right up until the moment an agent tries to actually transact. Every product page ships full Product plus Offer JSON-LD, prices sit in the raw HTML, Shop Pay and Apple Pay are wired to the cart, and the Shopify /cart/add endpoint is exposed for programmatic use. Then an agent hits the cart page and finds reCAPTCHA and hCaptcha waiting. The rest of the report is a variation on this theme.

The most consequential failure is check 4.2. Casper deploys both reCAPTCHA and hCaptcha on the cart and homepage, and the finding flags transaction_blocking: true. This is a four-point zero on a check that gates the entire agent-commerce use case. Every downstream investment in schema, wallets, and cart APIs is undercut by a challenge surface that no headless buyer will clear reliably. If Casper wants agent traffic to convert, this is the first thing to fix, and probably the only thing that matters this quarter.

Second, checkout structure and price transparency (checks 3.1 and 3.3) together bleed six points. The classifier couldn't even evaluate the checkout flow because the page requires an initialized cart to render, and https://casper.com/checkout returned a 403 during the scan. Shipping and tax are invisible pre-authorization; the cart page defers everything to "calculated at checkout." An agent quoting a total to a user has to guess, and guessing is what agents refuse to do. Publishing shipping thresholds and tax logic in a machine-readable form, even a simple JSON block, would recover most of this ground.

Third, the agent_native_content dimension is quietly the largest pool of lost points. Decision-support metadata (6.3) scored zero of three: none of the nine sampled products surface the kind of "who is this for" signal that lets an agent match a query like "cooling pillow for a hot side sleeper" to an SKU. Specification completeness (6.1) is uneven, with the Dream Hybrid mattress listing "QuickCool Cover, Premium Foam + Springs, Zoned Support" and no dimensions, firmness rating, or layer breakdown, while the SuperSoft Duvet Cover correctly lists 248 thread count and full size matrix. The gap is editorial, not technical.

Fourth, trust and policy gaps compound the picture. Check 4.4 finds Casper's terms and privacy pages are silent on automated agents, a two-point zero. Check 5.3 finds a "Last chance for big savings!" banner with a countdown timer displaying 00 HRS 00 MIN 00 SEC and no anchoring prior price, which the manipulation-tactics check (6.6) also flagged as a countdown_timer_no_date. Merchant verifiability (5.2) is missing a physical address and registration ID. None of these individually are large, but agents weighting trust will notice the pattern.

The bright spots are real: robots.txt allows all nine tracked crawlers including GPTBot, ClaudeBot, and PerplexityBot; llms.txt is published; Product plus Offer coverage is 9 of 9; and machine-native payment rails plus a Shopify cart endpoint mean the plumbing works. Casper has built an agent-ready catalog and then locked the front door.

The single highest-leverage move is removing reCAPTCHA and hCaptcha from the cart path, because until that happens, every other improvement on this report is theoretical.