StoreSteady Finding Taxonomy
StoreSteady's public vocabulary for the failure modes that make AI-driven commerce misread products, offers, policies, checkout paths, private catalog state, and validation evidence.
Relationship to the Breakage Index
These are StoreSteady's public failure categories. The Agentic Commerce Breakage Index will later measure them at corpus scale once detector coverage, sample size, and calibration gates are met.
Product truth and identity
Product truth
A product fact shown to shoppers, crawlers, channels, or AI systems conflicts with another public source for the same item. When that happens, the merchant lacks a single canonical product truth. Product truth is therefore the versioned, canonical, standards-based, time-stamped representation of a sellable item and its material commercial facts - identity, exact variant, price basis, availability, fulfillment constraints, and policy associations - that resolves consistently across every merchant-controlled surface and every agent-facing protocol.
Canonical product assertion
A single product fact (material, capacity, certification, compatibility, dimension, ingredient, warranty term) expressed in a machine-and-human-readable form, tied to a canonical source and timestamp, that serves as the atomic unit of `product-truth`.
Variant resolution
The process by which a system maps an offer, link, or answer to the exact purchasable variant (specific SKU, fully-specified color/size/configuration), not just to the parent product.
Offer interpretability and eligibility
Offer interpretability
A discount, price, availability, bundle, or shipping offer is hard for a crawler, shopping channel, or AI system to interpret from public evidence. Offer interpretability is therefore the degree to which a machine can unambiguously resolve a purchasable offer's identity, eligibility, current price, availability, shipping, returns, and required warnings from merchant-exposed data - structured markup, feeds, protocol responses, and rendered DOM.
Offer drift
Time-based divergence of price, availability, shipping, or policy facts across PDP, structured data, product feed, Merchant Center, and checkout cart, caused by asynchronous updates rather than a single point-in-time error.
Eligibility surface
The full set of machine-readable facts required for a system to determine whether a user or order qualifies for an offer, including region, account state, product restrictions, return-policy label, mandatory legal warnings, age/age-verification requirements, and quantity limits.
Agent-readable legal warning
A caution, restriction, or disclosure required to be presented before purchase (e.g., California Prop 65, age restriction, prescription requirement, recall notice, hazardous-material handling), exposed in a format an agent can reliably render or reason over.
Agentic checkout exceptions
Agentic checkout exception
Any failure, warning, or human-intervention condition raised during an agent-led checkout after the basket and purchase intent are already established, modeled at the business-logic level rather than as generic HTTP failure.
Checkout recoverability
The classification of a checkout exception by whether the agent can resolve it automatically, whether the buyer must provide input, or whether a human review is required.
Policy grounding and misstatement
AI policy misstatement
A shopping answer or channel interpretation says something about shipping, returns, warranty, privacy, or terms that the public policy pages do not support. Equivalently: a false, incomplete, or materially misleading statement by an AI system about a merchant's return, shipping, warranty, support, subscription, loyalty, or other transactional policy.
Policy-grounded answer
An answer about shipping, returns, warranty, or other transactional policies that is tied to authoritative policy artifacts and is auditable back to a cited source artifact and version.
Claim conflict and substantiation
Agent interaction and protocol identity
Agentic commerce
A commerce mode in which an AI agent can interpret user intent, discover and compare products or offers, apply constraints, and execute some or all purchase steps on the user's behalf, subject to explicit approval, merchant rules, and payment-network or protocol controls.
Consent checkpoint
A purchase step or risky action where explicit human confirmation is required before continuation, especially for payment authorization, submission of sensitive information, high-value orders, or first-time purchases from a new merchant.
Agent trust assertion
A cryptographic or protocol signal that claims an agent's identity, intent, or delegated authority and is verifiable by the merchant or payment network.
AI commerce QA discipline
AI commerce QA
The Test, Evaluation, Verification, and Validation (TEVV) discipline applied to commerce experiences consumed by AI systems, covering pre-deployment evaluation, field testing under deployment-like conditions, ongoing monitoring, red teaming, source verification, and regression tracking across discovery, policy answering, carting, checkout, and post-purchase.
AI mystery shopper
A StoreSteady scan pattern that checks what an AI or shopping channel appears to understand from the merchant surface. Operationalized as a mystery-shopping method in which trained humans, synthetic shoppers, or instrumented agents anonymously execute realistic buying or policy-inquiry scenarios to measure whether agent-facing commerce behavior matches merchant standards and consumer-facing promises.
Private exposure and access control
Post-fix validation and provenance
Post-fix validation
A finding needs a specific retest after the merchant changes the source page, markup, feed, policy, or checkout path. Post-fix validation is therefore the deployment-like re-verification performed after a remediation to prove the original issue no longer reproduces and that adjacent behaviors have not regressed across relevant channels, models, and protocols.
Provenance stamp
A versioned record linking a claim, answer, or artifact to the source snapshot, model version, transformation path, and timestamp that produced it.
Crawler access gap
Crawler access is covered in the published methodology standards layer. It does not get a separate V1 glossary expansion because this taxonomy focuses on agentic commerce failure primitives.
Run a scan against the taxonomy
StoreSteady maps scanner findings to public categories so operators can see what was detected, why it matters, and how the same evidence should be retested.
Run a free scan