Gaps. Discovery works, but trust does not.

Warby Parker scores 48 out of 100, landing squarely in the Gaps band, and the diagnosis is unusual. The infrastructure for agent commerce is largely in place: Stripe Link is detected across cart, checkout, and homepage, the Shopify cart endpoint is reachable, bot detection blocks nothing across 28 fetches, and CAPTCHAs are absent. What sinks the score is a content and metadata layer that assumes a human shopper with a browser, not a machine acting on a shopper's behalf. The largest single hole is specification completeness, where Warby Parker scores zero out of four. Only three of nine sampled product URLs surfaced structured specs at high signal. The Caldwell and Major detail pages list lens material, frame width, coatings, and UV blocking cleanly, but the Alcott and York pages return nothing parseable, and category pages like /eyeglasses/brown and /sunglasses/yellow-tinted-lens-colors expose prices without a single attribute an agent could match against a query like "wide titanium frames under 150 dollars." Decision-support metadata (6.3) and conversational descriptions (6.2) follow the same bimodal pattern: when the content team writes a real PDP, it is excellent, with width guides, prescription pricing tiers, and structured review breakdowns; when they do not, the page is essentially mute. Product attribute structure (2.2) is the next severe loss, scoring zero out of six. Across the sample, the only structured attribute consistently exposed is brand, and four of the nine URLs expose nothing at all. Inventory signals (2.3) score zero out of five for the same reason: only two URLs carry schema.org availability, and five carry neither schema nor text availability. The JSON-LD that does exist on pages like /sunglasses/major/cascades-crystal-with-polished-silver shows the team knows how to ship Offer markup with InStock and shipping details. They simply have not propagated it across the catalog. Transactability is the second front. Guest checkout (3.2) scores zero, with the cart presenting Sign In as the primary entry and no visible guest path. Price transparency preauth (3.3) also scores zero: no shipping or tax estimate is exposed before account creation. Combined with the empty cart blind spot on checkout flow (3.1), an agent attempting to complete a purchase has to create an account and commit a payment method before learning the total. That is a hard stop for autonomous buying. Smaller but telling: the sitemap at /sitemap.xml contains three locs, no product URLs, and no lastmod (1.3). The returns policy (5.1) is not linked from the homepage and /returns is unreachable, which costs three points and signals to agents that post-purchase risk is unbounded. Stated agent policy (4.4) is silent in the terms, leaving GPTBot, ClaudeBot, and PerplexityBot allowed in robots.txt but without any narrative permission an agent can cite. The wins are real. Price in raw HTML hits nine of nine (2.1), bot posture is clean (4.1, 4.2, 4.3), Stripe Link is present, the Shopify cart pattern is detected, and Bazaarvoice provides third-party trust anchoring. Warby Parker has built the rails. The biggest opportunity is to push the JSON-LD Product and Offer markup that already exists on Caldwell and Major across every PDP and category page, then expose a guest checkout path with shipping visible before sign-in.