Recorded run: our agent scored 100/100 · replay
It cleared every planted trap this time. A stronger baseline than most, but one run is not a guarantee; the mechanisms below still gate every agent purchase.
- [home] I found the Alpine Slip-On at $161.95 (p13), which is under the $189 all-in budget before taxes/fees. I need to open it to check if size 8 is available and see the final total. → open product
- [pdp:p13] Size 8 is available for the Alpine Slip-On at $161.95. I need to add it to cart to proceed toward checkout and verify the final all-in total. → add to cart
- [pdp:p13] Size 8 is available and already added to cart (badge shows 1 item). I need to view the cart to see the final total with any taxes or fees applied. → view cart
- [cart] Cart shows the Alpine Slip-On size 8 at $161.95 with subtotal matching; shipping is not calculated yet, so I need to proceed to checkout to see the final all-in total including taxes and fees. → begin checkout
- [checkout1] I'm on checkout step 1 of 3 (contact information page). I need to fill in email and name to proceed toward the final total with all taxes and fees applied. → submit contact
- [checkout2] Shipping step shows 'Calculated at payment' so I cannot see the final total yet; I need to proceed to checkout step 3 to see taxes and fees before placing the order. → choose shipping
- [checkout3] The item sold out but stock was relocated; the order summary still shows the Alpine Slip-On at $177.95 total (under budget), so I should attempt to place the order again. → place order×2
- [confirm] Order confirmed page shows the Alpine Slip-On size 8 was successfully purchased for $177.95 total, which is under the $189 budget. → finish
Narrative
Reformation reads well to agents on the content and infrastructure layers, but craters on the two things that actually let an agent transact: the checkout is gated behind reCAPTCHA on every surface, and product attributes are locked in prose instead of structured markup. A 63 is the accurate verdict for a store that describes its products beautifully but treats an autonomous buyer as a threat.
Where the points went
| Problem | Points lost | What an agent hits |
|---|---|---|
| Product attributes not structured | 6 | Agent must scrape "84% Lyocell and 16% Polyester" from prose to compare materials |
| reCAPTCHA on product, cart, checkout, homepage | 4 | Transaction blocked before payment rail ever loads |
| Checkout flow opaque to inspection | 4 | Agent cannot map steps or pre-fill without an initialized cart |
| Guest checkout absent | 4 | Forces account creation, killing one-shot purchase intent |
| Return policy not machine-readable | 3 | Agent cannot confirm 30-day window or conditions before recommending |
The evidence
- reCAPTCHA everywhere (Check 4.2, 0/4): Providers detected on all 10 sampled product URLs, plus cart, checkout, and homepage. Two Eli Oversized Shirt pages stack both reCAPTCHA and hCaptcha.
transaction_blocking: true. - Zero structured attributes (Check 2.2, 0/6): Across 10 products,
structuredHits: []andlabeledHits: []on every URL. The composition data exists in copy ("50% Polyester, 45% Rayon and 5% Nylon") but nothing is exposed as spec markup or a labeled table. - Checkout unreachable (Check 3.1, 2/6): Classifier could observe only one step, the empty bag state: "Your bag (0) Add some stuff to make your bag less empty. [CHECKOUT PAGE: unavailable]". Blind spot triggered on all three axes.
- No stated agent policy (Check 4.4, 0/2): Terms and privacy content returned nothing to classify. Agents defaulting to conservative behavior will assume automation is disallowed.
- No third-party trust anchors (Check 5.4, 0/2): Zero hits across Trustpilot, BBB, Yotpo, Bazaarvoice, and eight other platforms.
aggregate_rating_hits: 0.
What already works
- Crawler posture is fully open (6/6): GPTBot, ClaudeBot, PerplexityBot, Google-Extended, and Applebot-Extended all allowed in robots.txt.
- Product schema and pricing are clean (6/6 and 8/8): All 10 sampled URLs ship JSON-LD Product+Offer with prices in raw HTML.
- Content depth is elite (9/11 across dimension 6): High-signal specifications, conversational descriptions, and model measurements on 9 of 10 pages.
Fix first
- Remove reCAPTCHA from product and cart surfaces, or whitelist declared agent user-agents (+4 points).
- Emit product attributes as JSON-LD
additionalPropertyor a labeled spec table using the copy that already exists (+6 points). - Enable guest checkout and expose a linear, inspectable flow (+8 points combined across 3.1 and 3.2).
The single biggest lever is the CAPTCHA wall: the content and schema investment is already made, and it is being neutralized by a bot-detection posture that treats every agent as hostile.
ADiscoverability
Robots, sitemap, agent endpoints. strong.
CMachine Readability
Schema coverage, raw-HTML price, inventory, variants. partial.
CTransactability
Checkout flow, guest, transparency, payment rails, cart API. partial.
CAgent Posture
Bot detection, CAPTCHA, rate limiting, stated agent policy. partial.
ETrust Signals
Return policy, merchant verifiability, price history, third-party anchors. broken.
AAgent-Native Content
Specs, descriptions, decision support, delivery, stability, absence of manipulation. strong.