16 tags · the taxonomy is the spine of the benchmark

Failure taxonomy

Every failed journey gets a precise cause. These are the ways Indian D2C storefronts break AI shopping agents — each store's scorecard cites which of these it reproduced.

BOT_BLOCKED6 runs · 1 store

The storefront blocks automated browsers outright.

CAPTCHA4 runs · 2 stores

A CAPTCHA challenge gates progress.

OTP_WALL16 runs · 6 stores

A phone OTP is demanded before the journey can continue — unsatisfiable for an agent (and for anyone unwilling to hand over a number).

PAYMENT_HIDDEN

Payment options are hidden or unreachable.

NO_GUEST_CHECKOUT

Checkout requires creating an account.

LOGIN_WALL

An existing login is required to proceed.

APP_NUDGE_WALL

The site forces the journey into a native app.

BROKEN_FLOW17 runs · 3 stores

A step in the journey errors or dead-ends.

FORM_TRAP1 run

The checkout form rejects valid input or loops.

POPUP_INTERSTITIAL5 runs · 1 store

A popup or interstitial traps the flow.

PINCODE_GATE5 runs

A delivery-pincode gate blocks progress before intent.

UNSTRUCTURED_PDP1 run

The product page exposes no machine-readable structure.

SEARCH_BROKEN3 runs · 1 store

On-site search fails or returns unusable results.

OOS_TRAP

The scenario item is out of stock with no usable alternative path.

SLOW_LOAD

Pages take longer than 15 seconds to become usable.

AGENT_LIMITATION111 runs

The agent itself stalled (not a store fault) — the run is voided and excluded from store findings.

AGENT_LIMITATION (dashed, above) is a diagnostic, not a store finding: it marks runs voided because the agent itself stalled. A voided run is never published as a store failure and never deducts points on its own; voided runs are queued for re-runs.