CartPass · localhost:4001 · full reportScanned 19 Aug 2026

The verdict

Every AI shopper we sent bought from you.

Agent-ready our highest of four grades. Agents completed the tested purchase journey.

Rubric v1.0Harness 0.1.0Scan localhost-1787176735082Vertical general

We sent one AI shopper — the kind of software now buying on people’s behalf inside AI assistants — to buy “running shoes” from localhost:4001.

All 1 reached your payment screen with the right item in the basket.

The three scores

Three questions, three scores, never blended: Access (can agents get in, and how far), Journey (can agents who get in complete a purchase), Data-Layer (can agents understand what you sell). A blocked run’s Journey Score is “Not measurable”, never zero.

Anonymous agents

Access Score

100

Could agents get in, and how far. The scripted baseline is a control and is excluded from the composite.

Journey Score

100

Any agent profile completed: yes · every agent profile completed: yes.

Identified agents

Access Score

100

Could agents get in, and how far. The scripted baseline is a control and is excluded from the composite.

Journey Score

100

Any agent profile completed: yes · every agent profile completed: yes.

Data-Layer Score

Scan-level and identity-independent: what a machine can read from your pages without an agent ever visiting.

Headline finding — identified vs anonymous

Identified and anonymous agents scored the same Journey Score. Your stack does not differentiate on verifiable identity — the prerequisite for treating any verified agent differently from the anonymous long tail.

Anonymous agents

Agent	Attempts	Completed correctly	Mean credit	Spread	Furthest stage
baseline (scripted control)	1	1 of 1	100	0	S7 Payment reached
mock	1	1 of 1	100	0	S7 Payment reached

Identified agents

Agent	Attempts	Completed correctly	Mean credit	Spread	Furthest stage
baseline (scripted control)	1	1 of 1	100	0	S7 Payment reached
mock	1	1 of 1	100	0	S7 Payment reached

How far they got

Anonymous agents

1/1

Found the product

1/1

Chose size & colour

1/1

Added to basket

1/1

Reached checkout

1/1

Started guest checkout

1/1

Gave name & address

1/1

Reached payment

Identified agents

1/1

Found the product

1/1

Chose size & colour

1/1

Added to basket

1/1

Reached checkout

1/1

Started guest checkout

1/1

Gave name & address

1/1

Reached payment

We stop at the payment screen on purpose — no card details are ever touched.

What each agent did, attempt by attempt

Every attempt walks the seven stages with what the agent did, how long it took, the classified cause of any stop, and the evidence. Evidence paths are relative to this scan’s evidence directory.

Anonymous agents

baseline (scripted control)

01baseline · anonymousattempt baseline-unsigned-1-1787176738585 · CORRECT_COMPLETION

Passed S1 Discover, S2 Product page, S3 Add to cart, S4 Cart → checkout, S5 Checkout entry, S6 Customer info, S7 Payment reached — reached the payment screen with the cart contents verified.

Found the productS1 Discover
completed
navigated to the store home page0.2s
filled the search box with 'running shoes'0.4s
submitted the search0.2s
located the target product in the rendered catalog0.0s
search results
Evidence: screenshot (search results), html (search results)
Chose size & colourS2 Product page
completed
opened the product page0.2s
confirmed the product page for 'Trailblazer GTX Trail Shoe'0.0s
selected size '7'0.4s
selected colour 'Black'0.0s
product confirmed and required variants selected0.0s
product page with variants selected
Evidence: screenshot (product page with variants selected), html (product page with variants selected)
Added to basketS3 Add to cart
completed
clicked add to cart0.3s
opened the cart0.0s
cart verified against the task target0.9s
cart contents
Evidence: screenshot (cart contents), html (cart contents), json (parsed cart line items + cross-check)
Reached checkoutS4 Cart → checkout
completed
transitioned from cart to checkout0.3s
reached the checkout entry point0.0s
checkout entry
Evidence: screenshot (checkout entry), html (checkout entry)
Started guest checkoutS5 Checkout entry
completed
entered checkout via the guest path0.3s
guest checkout entered0.0s
customer info form
Evidence: screenshot (customer info form), html (customer info form)
Gave name & addressS6 Customer info
completed
filled the contact and shipping form with synthetic details1.7s
submitted the customer information form0.4s
customer information accepted0.0s
post-customer-info
Evidence: screenshot (post-customer-info), html (post-customer-info)
Reached paymentS7 Payment reached
completed
payment method UI reached through checkout — journey stops here, no payment input is ever touched0.0s
payment screen
Evidence: screenshot (payment screen), html (payment screen), url_trail (URLs visited during the attempt), network (HTTP responses observed during the attempt)

mock

01mock · anonymousattempt mock-unsigned-1-1787176744331 · CORRECT_COMPLETION

Passed S1 Discover, S2 Product page, S3 Add to cart, S4 Cart → checkout, S5 Checkout entry, S6 Customer info, S7 Payment reached — reached the payment screen with the cart contents verified.

Found the productS1 Discover
completed
navigated to the store home page0.3s
filled searchbox "Search" with 'running shoes' — searching the rendered catalog for 'running shoes'0.3s
clicked button "Search" — running the search0.3s
clicked link "Trailblazer GTX Trail Shoe" — opening the product page for 'Trailblazer GTX Trail Shoe'0.5s
located 'Trailblazer GTX Trail Shoe' in the rendered catalog0.0s
product located
Evidence: json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), screenshot (product located), html (product located)
Chose size & colourS2 Product page
completed
chose '7' in combobox "Size" — choosing Size '7'0.2s
chose 'Black' in combobox "Colour" — choosing Colour 'Black'0.2s
confirmed 'Trailblazer GTX Trail Shoe' and selected Size '7', Colour 'Black'0.0s
product page with variants selected
Evidence: json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), screenshot (product page with variants selected), html (product page with variants selected)
Added to basketS3 Add to cart
completed
clicked button "Add to cart" — adding the selected variant to the cart0.3s
cart verified against the task target0.8s
cart contents
Evidence: json (what the agent saw and the action it chose), screenshot (cart contents), html (cart contents), json (parsed cart line items + cross-check)
Reached checkoutS4 Cart → checkout
completed
clicked button "Checkout" — the cart holds the target item — moving to checkout0.3s
transitioned from cart to checkout0.0s
checkout entry
Evidence: json (what the agent saw and the action it chose), screenshot (checkout entry), html (checkout entry)
Started guest checkoutS5 Checkout entry
completed
clicked button "Continue as guest" — entering checkout as a guest0.3s
entered checkout without an account and reached the customer information form0.0s
customer info form
Evidence: json (what the agent saw and the action it chose), screenshot (customer info form), html (customer info form)
Gave name & addressS6 Customer info
completed
filled textbox "Full name" with 'Test Agentbuyer' — entering the synthetic name into 'Full name'0.4s
filled textbox "Email" with 'scan-test@example.invalid' — entering the synthetic email into 'Email'0.2s
filled textbox "Phone" with '+44 7700 900123' — entering the synthetic phone into 'Phone'0.3s
filled textbox "Address line 1" with '1 Test Street' — entering the synthetic address1 into 'Address line 1'0.3s
filled textbox "City" with 'London' — entering the synthetic city into 'City'0.3s
filled textbox "Postcode" with 'SW1A 1AA' — entering the synthetic postcode into 'Postcode'0.2s
filled textbox "Country" with 'GB' — entering the synthetic country into 'Country'0.3s
clicked button "Continue to payment" — submitting the contact and shipping details0.5s
customer information accepted0.0s
post-customer-info
Evidence: json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), screenshot (post-customer-info), html (post-customer-info)
Reached paymentS7 Payment reached
completed
payment method UI reached through checkout — journey stops here, no payment input is ever touched0.0s
payment screen
Evidence: screenshot (payment screen), html (payment screen), url_trail (URLs visited during the attempt), network (HTTP responses observed during the attempt)

Identified agents

baseline (scripted control)

01baseline · identifiedattempt baseline-signed-1-1787176750545 · CORRECT_COMPLETION

Passed S1 Discover, S2 Product page, S3 Add to cart, S4 Cart → checkout, S5 Checkout entry, S6 Customer info, S7 Payment reached — reached the payment screen with the cart contents verified.

Found the productS1 Discover
completed
navigated to the store home page0.2s
filled the search box with 'running shoes'0.4s
submitted the search0.4s
located the target product in the rendered catalog0.0s
search results
Evidence: screenshot (search results), html (search results)
Chose size & colourS2 Product page
completed
opened the product page0.2s
confirmed the product page for 'Trailblazer GTX Trail Shoe'0.0s
selected size '7'0.5s
selected colour 'Black'0.0s
product confirmed and required variants selected0.0s
product page with variants selected
Evidence: screenshot (product page with variants selected), html (product page with variants selected)
Added to basketS3 Add to cart
completed
clicked add to cart0.2s
opened the cart0.0s
cart verified against the task target0.7s
cart contents
Evidence: screenshot (cart contents), html (cart contents), json (parsed cart line items + cross-check)
Reached checkoutS4 Cart → checkout
completed
transitioned from cart to checkout0.5s
reached the checkout entry point0.0s
checkout entry
Evidence: screenshot (checkout entry), html (checkout entry)
Started guest checkoutS5 Checkout entry
completed
entered checkout via the guest path0.4s
guest checkout entered0.0s
customer info form
Evidence: screenshot (customer info form), html (customer info form)
Gave name & addressS6 Customer info
completed
filled the contact and shipping form with synthetic details1.9s
submitted the customer information form0.4s
customer information accepted0.0s
post-customer-info
Evidence: screenshot (post-customer-info), html (post-customer-info)
Reached paymentS7 Payment reached
completed
payment method UI reached through checkout — journey stops here, no payment input is ever touched0.0s
payment screen
Evidence: screenshot (payment screen), html (payment screen), url_trail (URLs visited during the attempt), network (HTTP responses observed during the attempt)

mock

01mock · identifiedattempt mock-signed-1-1787176756579 · CORRECT_COMPLETION

Passed S1 Discover, S2 Product page, S3 Add to cart, S4 Cart → checkout, S5 Checkout entry, S6 Customer info, S7 Payment reached — reached the payment screen with the cart contents verified.

Found the productS1 Discover
completed
navigated to the store home page0.4s
filled searchbox "Search" with 'running shoes' — searching the rendered catalog for 'running shoes'0.4s
clicked button "Search" — running the search0.5s
clicked link "Trailblazer GTX Trail Shoe" — opening the product page for 'Trailblazer GTX Trail Shoe'0.3s
located 'Trailblazer GTX Trail Shoe' in the rendered catalog0.0s
product located
Evidence: json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), screenshot (product located), html (product located)
Chose size & colourS2 Product page
completed
chose '7' in combobox "Size" — choosing Size '7'0.3s
chose 'Black' in combobox "Colour" — choosing Colour 'Black'0.3s
confirmed 'Trailblazer GTX Trail Shoe' and selected Size '7', Colour 'Black'0.0s
product page with variants selected
Evidence: json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), screenshot (product page with variants selected), html (product page with variants selected)
Added to basketS3 Add to cart
completed
clicked button "Add to cart" — adding the selected variant to the cart0.2s
cart verified against the task target0.8s
cart contents
Evidence: json (what the agent saw and the action it chose), screenshot (cart contents), html (cart contents), json (parsed cart line items + cross-check)
Reached checkoutS4 Cart → checkout
completed
clicked button "Checkout" — the cart holds the target item — moving to checkout0.3s
transitioned from cart to checkout0.0s
checkout entry
Evidence: json (what the agent saw and the action it chose), screenshot (checkout entry), html (checkout entry)
Started guest checkoutS5 Checkout entry
completed
clicked button "Continue as guest" — entering checkout as a guest0.2s
entered checkout without an account and reached the customer information form0.0s
customer info form
Evidence: json (what the agent saw and the action it chose), screenshot (customer info form), html (customer info form)
Gave name & addressS6 Customer info
completed
filled textbox "Full name" with 'Test Agentbuyer' — entering the synthetic name into 'Full name'0.2s
filled textbox "Email" with 'scan-test@example.invalid' — entering the synthetic email into 'Email'0.1s
filled textbox "Phone" with '+44 7700 900123' — entering the synthetic phone into 'Phone'0.2s
filled textbox "Address line 1" with '1 Test Street' — entering the synthetic address1 into 'Address line 1'0.2s
filled textbox "City" with 'London' — entering the synthetic city into 'City'0.4s
filled textbox "Postcode" with 'SW1A 1AA' — entering the synthetic postcode into 'Postcode'0.3s
filled textbox "Country" with 'GB' — entering the synthetic country into 'Country'0.2s
clicked button "Continue to payment" — submitting the contact and shipping details0.4s
customer information accepted0.0s
post-customer-info
Evidence: json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), json (what the agent saw and the action it chose), screenshot (post-customer-info), html (post-customer-info)
Reached paymentS7 Payment reached
completed
payment method UI reached through checkout — journey stops here, no payment input is ever touched0.0s
payment screen
Evidence: screenshot (payment screen), html (payment screen), url_trail (URLs visited during the attempt), network (HTTP responses observed during the attempt)

Failure taxonomy

No stage failures were recorded in the tested stages.

Remediation roadmap

No failure codes were observed in the tested stages, so there is nothing to remediate from this scan.

Can AI read your shelves?

Every check, its published point value, and what we found. The sub-score is never a black box — you can see exactly which check failed and what fixing it earns.

Prices and photos — readable. AI can quote you accurately.

Structured product data quality (schema.org Product / Offer) — 3 product page(s) examined; scored on populated required properties, not mere presence. /product/trailblazer-gtx: 5/5 required properties, 15 offer(s): name ok (Trailblazer GTX Trail Shoe), price ok (129), priceCurrency ok (GBP), availability ok (InStock), identifier ok (sku=FX-TRB-100) (1 ld+json block). /product/pacer-lite: 5/5 required properties, 12 offer(s): name ok (Pacer Lite Road Shoe), price ok (89), priceCurrency ok (GBP), availability ok (InStock), identifier ok (sku=FX-PCR-200) (1 ld+json block). /product/summit-max: 5/5 required properties, 12 offer(s): name ok (Summit Max Cushion Shoe), price ok (145), priceCurrency ok (GBP), availability ok (InStock), identifier ok (sku=FX-SMT-300) (1 ld+json block). Product pages (3) from: sitemap http://localhost:4001/sitemap.xml.

Evidence: text (GET http://localhost:4001/product/trailblazer-gtx -> 200 (product page)), text (GET http://localhost:4001/product/pacer-lite -> 200 (product page)), text (GET http://localhost:4001/product/summit-max -> 200 (product page))

24/24

Sizes and colours — published per variant. AI knows what you stock.

Variant-level machine readability — Can an agent resolve each variant's price and availability without operating the UI? /product/trailblazer-gtx: 15 variants via multiple offers: 15/15 carry distinguishing attributes (label=Trailblazer GTX Trail Shoe — Size 7, Black; label=Trailblazer GTX Trail Shoe — Size 7, Cobalt; label=Trailblazer GTX Trail Shoe — Size 7, Sand), 15/15 carry both their own price and availability. /product/pacer-lite: 12 variants via multiple offers: 12/12 carry distinguishing attributes (label=Pacer Lite Road Shoe — Size 6, White; label=Pacer Lite Road Shoe — Size 6, Coral; label=Pacer Lite Road Shoe — Size 7, White), 12/12 carry both their own price and availability. /product/summit-max: 12 variants via multiple offers: 12/12 carry distinguishing attributes (label=Summit Max Cushion Shoe — Size 7, Slate; label=Summit Max Cushion Shoe — Size 7, Lime; label=Summit Max Cushion Shoe — Size 8, Slate), 12/12 carry both their own price and availability. Product pages (3) from: sitemap http://localhost:4001/sitemap.xml.

12/12

Live prices — match the page a shopper sees.

Price / inventory freshness consistency — Structured price vs rendered page (all-or-nothing across 3 page(s)). /product/trailblazer-gtx: JSON-LD price 129 appears in the rendered page as "129.00". /product/pacer-lite: JSON-LD price 89 appears in the rendered page as "89.00". /product/summit-max: JSON-LD price 145 appears in the rendered page as "145.00". Product pages (3) from: sitemap http://localhost:4001/sitemap.xml.

10/10

Guest checkout — easy to find without an account.

Guest checkout detectable — Heuristic (plain fetch, no browser): probed /cart, /checkout for guest-checkout wording/ids, forced-login markers and login redirects. /cart: no guest signal, no login marker, no checkout entry point; /checkout: guest signal "check out as a guest", no login marker, checkout entry point present. Verdict: explicit guest-checkout affordance on /checkout (check out as a guest), with no forced-login marker. Cross-checked against journey evidence in the full report.

Evidence: text (GET http://localhost:4001/cart -> 200 (purchase path)), text (GET http://localhost:4001/checkout -> 200 (purchase path))

8/8

Delivery & returns — invisible. AI can’t answer “when will it arrive?”

Shipping, tax and returns policy machine readability — Shipping/tax and returns policies discoverable by an agent, plus schema.org policy markup (0/3). shipping/tax: no link and no page at /policies/shipping-policy -> 404, /shipping -> 404; returns/refunds: no link and no page at /policies/refund-policy -> 404, /returns -> 404; structured markup: no OfferShippingDetails or MerchantReturnPolicy in product JSON-LD.

Evidence: text (GET http://localhost:4001/ -> 200 (homepage)), text (GET http://localhost:4001/policies/shipping-policy -> 404 (policy probe)), text (GET http://localhost:4001/shipping -> 404 (policy probe)), text (GET http://localhost:4001/policies/refund-policy -> 404 (policy probe)), text (GET http://localhost:4001/returns -> 404 (policy probe)), text (GET http://localhost:4001/product/trailblazer-gtx -> 200 (product page)), text (GET http://localhost:4001/product/pacer-lite -> 200 (product page)), text (GET http://localhost:4001/product/summit-max -> 200 (product page))

0/6

Verified-agent check-in — missing. You can’t tell a trusted agent from a scraper, so both get the same treatment.

Web Bot Auth key directory — Web Bot Auth (RFC 9421) key directory: GET /.well-known/http-message-signatures-directory -> HTTP 404. No key directory, so a signed agent has no published key to verify against.

Evidence: text (GET http://localhost:4001/.well-known/http-message-signatures-directory -> 404 (Web Bot Auth key directory))

0/6

Agent checkout standard (UCP) — not published. Agents have to click through your site like a person.

UCP well-known endpoint — UCP (Google/Shopify) discovery+cart surface: GET /.well-known/ucp -> HTTP 404. Not present. Shopify auto-enrolls merchants via Agentic Storefronts, so absence on a Shopify store is itself a finding.

Evidence: text (GET http://localhost:4001/.well-known/ucp -> 404 (UCP well-known))

0/6

Agent checkout standard (ACP) — not published. There is no direct route for an agent purchase.

ACP endpoint / product-feed hints — ACP (OpenAI/Stripe) in-chat checkout surface. Probed: GET /.well-known/acp -> HTTP 404; homepage <link>/<meta> tags mentioning ACP/agentic checkout/product feed: none; sitemap entries that look like agent product feeds: none; response headers: none. No ACP endpoint and no feed/link hints detected.

Evidence: text (GET http://localhost:4001/.well-known/acp -> 404 (ACP well-known)), text (GET http://localhost:4001/ -> 200 (homepage))

0/6

Direct connection for AI assistants — not advertised.

MCP endpoint — MCP tool/endpoint access. Probed: GET /.well-known/mcp -> HTTP 404; homepage link/meta tags advertising an MCP endpoint: none; response headers: none. No MCP endpoint detected.

Evidence: text (GET http://localhost:4001/.well-known/mcp -> 404 (MCP well-known)), text (GET http://localhost:4001/ -> 200 (homepage))

0/4

Agent payment signals — absent. Agent-initiated payments have no fast path.

AP2 / Shared Payment Token signals — AP2 (FIDO-governed payment authorization) and Shared Payment Token (Stripe/ACP). Probed: GET /.well-known/ap2 -> HTTP 404; homepage link/meta tags: none; response headers: none. No signal detected. AP2 and Shared Payment Token support is largely negotiated inside checkout, so a zero here means "not statically observable", not "definitely unsupported" — the paid battery drives this live.

Evidence: text (GET http://localhost:4001/.well-known/ap2 -> 404 (AP2 well-known)), text (GET http://localhost:4001/ -> 200 (homepage))

0/4

Card-network agent programmes — not signalled.

Card-network agent programme signals — Visa TAP / Mastercard Agent Pay. Inspected homepage link/meta tags and response headers for card-network agent-program markers (no probe: neither program publishes a well-known path). No signal detected. Visa Trusted Agent Protocol and Mastercard Agent Pay are enrolled at the acquirer/PSP layer with no public discovery document, so this is an honest zero rather than a judgment about the merchant's enrolment.

Evidence: text (GET http://localhost:4001/ -> 200 (homepage))

0/4

AI guide file — published. It tells an AI assistant where to look.

llms.txt — llms.txt agent guidance: GET /llms.txt -> HTTP 200, body is not JSON. Present, 1364 bytes of text.

Evidence: text (GET http://localhost:4001/llms.txt -> 200 (llms.txt))

3/3

Rules for AI crawlers — not spelled out. AI crawlers have to guess what you allow.partial

robots.txt AI-agent directives — robots.txt parses: 1 User-agent group(s) including a "*" group, 1 Sitemap line(s). No AI-agent user-agent is addressed. None of gptbot, chatgpt-user, oai-searchbot, claudebot, claude-user, claude-searchbot (and 15 others we look for) appear, so AI crawlers can only fall back to the generic rules and guess at your intent.

Evidence: text (GET http://localhost:4001/robots.txt -> 200 (robots.txt))

2/4

Product sitemap — valid. AI can find every product.

Sitemap validity — Sitemap (declared in robots.txt) http://localhost:4001/sitemap.xml: <urlset> with 10 <loc> entries. Parses as XML with at least one <loc> — usable for product discovery.

Evidence: text (GET http://localhost:4001/sitemap.xml -> 200 (sitemap))

3/3

Methodology, versioning and limitations

What this scan pinned

Field	Value
Scan ID	localhost-1787176735082
Scan date	2026-08-19T21:59:22.561Z
Rubric version	v1.0
Harness version	0.1.0
Agent profiles and model IDs	baseline (scripted); mock (llm, model mock)
Identity modes tested	anonymous, identified
Task query	running shoes
Pinned product	not pinned
Vertical profile	general
Valid runs scored	4

Scores are only comparable within a rubric version and vertical cohort. Weights are declared priors, published in advance and re-tuned as a new rubric version — never silently changed.

Identity modes

Identified runs use a cryptographically verifiable identity with an unknown operator. Allowlists key off operator identity, not signature presence, so the identified result is a lower bound on how this site treats agents it has chosen to trust.
Anonymous runs represent the long tail of unverified agents, which bot defenses may treat more strictly than agents a merchant has chosen to trust.
The identified/anonymous delta measures whether your stack differentiates on verifiable identity at all — the prerequisite for treating any verified agent well.

Guardrails

The journey stops the moment the payment screen is reached. No input ever touches card fields or payment iframes, no account is created, and no CAPTCHA is solved or evaded.

Filtering our traffic out of your analytics

Scanning leaves traces in your systems and we would rather you could remove them. Every synthetic identity uses one email domain we own, and every request carries an honest user-agent naming the product and a contact URL. One rule each:

Analytics and session tools — exclude sessions whose user-agent contains CartPassBot. Full string for this scan: CartPassBot/0.1.0 (+https://cartpass.ai/contact)
Abandoned-cart flows, CRM and remarketing lists — exclude any contact whose email address ends in @example.invalid. Consented journey runs will trigger abandoned-cart email to those addresses; we receive it, never engage with it, and delete it.

Known limitations

Results reflect the tested identity mode. Anonymous agents may be treated more strictly by bot defenses than verified production agents, and the identified mode measures identity differentiation, not any specific commercial agent’s experience.
Unsolicited scans are robots.txt-compliant and therefore measure only part of the journey on many platforms; full-journey results exist only for consented scans.
LLM agents are stochastic. We report completion rates and spread, but small run counts carry wider error bars.
Our agent profiles approximate, but are not, the commercial agent products. A quarterly validation battery measures the drift and publishes it in the methodology changelog.
Failure distributions differ by vertical (fashion, electronics, grocery, digital goods). Every scan records a vertical profile so results can be cohorted; scores are only comparable within a rubric version.