Live Lab · public web only

Two independent public-web references, measured on two agent surfaces.

These are provisional readings, not final Commerce4 scores. These public outputs are built without using private store access or analytics and contain no executed UCP or WebMCP results for either storefront. Every finding below carries an explicit evidence level, and any capability that requires private or active testing is scored conservatively and marked pending.

Relationship neutrality

Independent public-web reference. Inclusion does not state or imply customer status, partnership, endorsement, participation, authorization, or private access.

Named profiles use only independently observable public-web information. No private, API, account or relationship data appears anywhere in these outputs.

Private validation access: NOT CONNECTED for both references. No merchant-authorized credential, admin, analytics or private connection is used, so every transactional and attribution item stays NEEDS_STORE_ACCESS. Scores are unaffected by this note.

Provisional — public web only

Two surfaces, two scores

UCP · server-to-server is the protocol layer: catalog, cart and delegated checkout between machines. WebMCP · browser session is what an agent can actually do inside the buyer’s session, now that Shopify exposes storefront tools on Liquid themes by default. A store can score well on one and poorly on the other, so we never collapse them into a single number.

Benchmark / positive control

Zinus México

zinus.mx
Independent · provisional · public web only

UCP · server-to-server

70/ 100

Competitive (60–79)

WebMCP · browser session

61/ 100

Competitive (60–79)

Publicly identifies as Shopify

Stress test / complexity control

Photobook México

photobookmexico.mx
Independent · provisional · public web only

UCP · server-to-server

53/ 100

Emerging (40–59)

WebMCP · browser session

50/ 100

Emerging (40–59)

Publicly identifies as Shopify

Buyer-agent eval matrix (2 stores × 2 surfaces)

25 UCP evals per store plus a WebMCP subset, with six metrics recorded separately instead of one pass/fail. Rates below only count cases that have been scored from public evidence; everything transactional reads “pending” until the harness runs.

MetricZinus MéxicoUCPZinus MéxicoWebMCPPhotobook MéxicoUCPPhotobook MéxicoWebMCP
Discovery success76%50%59%50%
Correct product80%pending58%pending
Correct variant / configuration50%pending17%pending
Browser-agent action successpendingpending
Cart fidelitypendingpendingpendingpending
Checkout readinesspendingpendingpendingpending

Surface detail

Zinus México

WebMCP · browser session · 8 eval cases

MetricScored rateCases
Discovery success50%1 scored · 7 pending
Correct productpending0 scored · 8 pending
Correct variant / configurationpending0 scored · 7 pending · 1 n/a
Browser-agent action successpending0 scored · 8 pending
Cart fidelitypending0 scored · 8 pending
Checkout readinesspending0 scored · 8 pending
  • Tool exposure & discovery

    15 / 20
    Inferred

    Liquid storefront, so the platform-default WebMCP tool set (catalog search, product, cart, checkout, policies) is expected to be present without merchant configuration.

  • Product & variant resolution

    13 / 20
    Inferred

    Fixed SKUs with conventional size and height options make in-session resolution the least risky part of the flow.

  • Cart action fidelity

    13 / 25
    Needs store access

    Add-to-cart fidelity for size and quantity is plausible but unverified; these products carry no per-item configuration, so no line attributes are required.

  • Checkout progression

    10 / 20
    Needs store access

    Checkout progression, totals and shipping truth in an agent-driven session have not been exercised.

  • Policy & support answerability

    10 / 15
    Verified public

    Warranty, returns and certification content is published and differentiated by product type, which is what the policy tool answers from.

Representative eval cases

  • zinus_mx_webmcp_01Inferred

    Discover which storefront tools this page exposes to the agent

    Tool exposure & discoveryCatalog search, product lookup, cart, checkout, policy tools discoverable

    Shopify enabled WebMCP on Liquid storefronts; agent support is still Chromium origin-trial limited.

    Discovery successPartialCorrect productPending runCorrect variant / configurationPending runBrowser-agent action successPending runCart fidelityPending runCheckout readinessPending run
  • zinus_mx_webmcp_02Needs store access

    Search the catalog in session for 'colchón queen memory foam'

    In-session catalog searchSearch tool returns the correct product set

    Discovery successPending runCorrect productPending runCorrect variant / configurationPending runBrowser-agent action successPending runCart fidelityPending runCheckout readinessPending run
  • zinus_mx_webmcp_03Needs store access

    Select the queen variant on the product page

    Variant selection in sessionSelected variant matches the requested size

    Discovery successPending runCorrect productPending runCorrect variant / configurationPending runBrowser-agent action successPending runCart fidelityPending runCheckout readinessPending run
  • zinus_mx_webmcp_04Needs store access

    Add the selected variant to cart via the storefront action

    Cart action fidelityCart line matches variant and quantity exactly

    Discovery successPending runCorrect productPending runCorrect variant / configurationPending runBrowser-agent action successPending runCart fidelityPending runCheckout readinessPending run
  • zinus_mx_webmcp_05Needs store access

    Increase quantity to two from inside the cart

    Cart mutation in sessionQuantity change reflected without losing the line

    Discovery successPending runCorrect productPending runCorrect variant / configurationPending runBrowser-agent action successPending runCart fidelityPending runCheckout readinessPending run
  • zinus_mx_webmcp_06Needs store access

    Ask the shipping policy question without leaving the page

    Policy answerabilityPolicy tool answers from published terms

    Discovery successPending runCorrect productPending runCorrect variant / configurationN/ABrowser-agent action successPending runCart fidelityPending runCheckout readinessPending run

Photobook México

WebMCP · browser session · 12 eval cases

MetricScored rateCases
Discovery success50%1 scored · 11 pending
Correct productpending0 scored · 12 pending
Correct variant / configurationpending0 scored · 11 pending · 1 n/a
Browser-agent action successpending0 scored · 12 pending
Cart fidelitypending0 scored · 12 pending
Checkout readinesspending0 scored · 12 pending
  • Tool exposure & discovery

    14 / 20
    Inferred

    Platform-default WebMCP tools are expected, but the configuration builder is the surface that matters and may sit outside standard storefront actions.

  • Product & variant resolution

    8 / 20
    Needs store access

    The purchasable unit is a configuration rather than a SKU, so 'correct variant' is only meaningful once the builder state is readable by the agent.

  • Cart action fidelity

    9 / 25
    Needs store access

    Two separate unknowns: (a) whether size, material, page count and instructions survive as LINE attributes and can be read back from the cart line after an edit, and (b) whether cart-wide metadata updates are observable through the cart-attribute update/change event.

  • Checkout progression

    9 / 20
    Needs store access

    Cross-border totals, duties and configuration persistence at checkout are all unverified.

  • Policy & support answerability

    10 / 15
    Verified public

    Personalized-goods return terms and international shipping content are published, giving the policy tool something truthful to answer from.

Representative eval cases

  • photobook_mx_webmcp_01Inferred

    Discover the storefront tools exposed on a configurable product page

    Tool exposure & discoveryCatalog, product, cart, checkout and policy tools discoverable in session

    WebMCP is on by default for Liquid storefronts; agent-side support remains Chromium origin-trial limited.

    Discovery successPartialCorrect productPending runCorrect variant / configurationPending runBrowser-agent action successPending runCart fidelityPending runCheckout readinessPending run
  • photobook_mx_webmcp_02Needs store access

    Build a 20x20 cm hardcover, 40-page configuration in session

    Configuration → attributesEvery chosen option captured as structured context

    Discovery successPending runCorrect productPending runCorrect variant / configurationPending runBrowser-agent action successPending runCart fidelityPending runCheckout readinessPending run
  • photobook_mx_webmcp_03Needs store access

    Add the configured book to cart and read back the line attributes

    Line attributesSize, material, page count and design reference present as attributes on that cart LINE

    Per-item configuration belongs on the cart LINE, not on the cart. The Shopify cart-attribute update/change event applies to cart-wide metadata only.

    Discovery successPending runCorrect productPending runCorrect variant / configurationPending runBrowser-agent action successPending runCart fidelityPending runCheckout readinessPending run
  • photobook_mx_webmcp_04Needs store access

    Change the cover material and re-read the cart line attributes

    Line-attribute round-tripThe updated material is present when the cart line is read back; verification is by read-back, not by any cart-attribute event

    The Shopify cart-attribute update/change event covers cart-wide metadata only and does not apply to line attributes.

    Discovery successPending runCorrect productPending runCorrect variant / configurationPending runBrowser-agent action successPending runCart fidelityPending runCheckout readinessPending run
  • photobook_mx_webmcp_05Needs store access

    Attach printing instructions as a line attribute

    Free-text context fidelityInstruction text survives verbatim on the line

    Discovery successPending runCorrect productPending runCorrect variant / configurationPending runBrowser-agent action successPending runCart fidelityPending runCheckout readinessPending run
  • photobook_mx_webmcp_06Needs store access

    Add two identical 20x20 hardcovers with different engraving text

    Line separationSame variant with different line attributes stays as two separate cart lines

    Discovery successPending runCorrect productPending runCorrect variant / configurationPending runBrowser-agent action successPending runCart fidelityPending runCheckout readinessPending run

Attribute fidelity

Does the configuration survive from builder to checkout?

Commerce4 separates line attributes (per-item configuration: engraving, size, material, page count, design reference, instructions) from cart-level attributes (order-wide metadata: gift message, order note, agent provenance).

Line attributes

  • Line attributes are the decisive mechanism: size, material, page count, design reference and printing instructions are per-item configuration and must travel on the cart line.
  • Configuration fidelity must be verified on line attributes; a single cart-level attribute cannot faithfully represent each configured item in a multi-item order.
  • Two identical variants with different line attributes must remain separate cart lines — merging them silently destroys the buyer's intent.
  • If configuration lives only in builder state and never lands on the line, an agent can complete a purchase that does not match what the buyer asked for.

Cart-level attributes

  • Cart-level attributes stay narrow here: gift message, order note and agent provenance — order-wide metadata, not configuration.
  • The cart-attribute change event remains useful for verification of that order-wide metadata.

Zinus México — browser findings

  1. 01Needs WorkInferred

    WebMCP tools present by platform default, not by merchant design

    Because the storefront runs on Liquid, the standard agent tool set should be exposed without any merchant action. That is a floor, not a strategy: nothing here is tuned for agent operation.

    Recommended fix
    Confirm the exposed tool list in a Chromium origin-trial session and record which tools answer usefully for this catalog.

  2. 02Needs WorkInferred

    Variant selection is the main in-session failure candidate

    Size is well structured, but height and firmness read as prose. An agent operating the page may add the right model in the wrong configuration.

    Recommended fix
    Normalize height and firmness into explicit selectable options so the browser agent selects rather than infers.

  3. 03CriticalNeeds store access

    Agent provenance is not captured in the cart

    Without an attribute marking agent-initiated sessions, browser-agent revenue cannot be separated from ordinary traffic later.

    Recommended fix
    Write an agent-provenance cart-level attribute on agent-initiated sessions and carry it through to the order.

  4. 04StrongVerified public

    Policy answers are the strongest in-session surface

    Warranty differentiation by product type and public certification give the policy tool accurate grounding, which reduces hallucinated support answers.

    Recommended fix
    Keep policy pages canonical and avoid moving terms into images or PDFs.

  5. 05Needs WorkNeeds store access

    Out-of-stock handling in session is unverified

    If an unavailable variant fails silently, the agent reports success on a cart that cannot convert.

    Recommended fix
    Test unavailable-variant paths and ensure the action returns an explicit error.

Photobook México — browser findings

  1. 01CriticalNeeds store access

    Configuration fidelity is the single highest-risk agent failure

    The purchasable unit is a configuration. An agent that adds the base product without carrying every chosen option as a LINE attribute produces a wrong order that still looks like a successful checkout.

    Recommended fix
    Persist every configuration choice as an explicit line attribute at add-to-cart time, keep identical variants with different line attributes as separate lines, and expose them back for verification.

  2. 02CriticalNeeds store access

    Cart-level attribute change event is not known to be used

    The cart-attribute update/change event covers order-wide metadata only — gift message, order note, agent provenance. Without emitting and consuming it, an agent editing that cart-wide metadata cannot confirm the edit landed before proceeding to checkout. Line-attribute edits are outside this event's scope and must be verified by reading the cart line back.

    Recommended fix
    Emit cart-wide metadata updates through the standard storefront cart-attribute event so any agent can observe them, and separately expose line attributes for read-back verification.

  3. 03Needs WorkNeeds store access

    Free-text instructions may not survive to checkout

    Printing instructions and design notes are the most common context to be dropped between builder, cart and order.

    Recommended fix
    Round-trip instruction text through line attributes and assert it verbatim at checkout.

  4. 04Needs WorkInferred

    Locale and destination switching can reset context

    The store sells internationally with a large region selector; a mid-session destination change is a plausible point where configuration or totals desynchronise.

    Recommended fix
    Verify that changing destination preserves line and cart attributes and updates duties and totals coherently.

  5. 05Needs WorkVerified public

    Made-to-order availability reads as stock

    Stock and add-to-cart signals contradict each other in places, which an in-session agent will interpret literally.

    Recommended fix
    Signal made-to-order production explicitly instead of borrowing inventory semantics.

Category comparison

Same seven weighted categories as the full methodology. Bars show provisional points earned out of the category weight.

Catalog & Product Semantics

weight 25 pts

Zinus México

20 / 25
Verified public

Product pages publicly expose titles, SKU-like identifiers, price, height and size choices, and detailed descriptions.

Photobook México

16 / 25
Verified public

Catalog exposes photobook families (Spanish, American/hardcover, Italian/softcover), packages and other products, but configuration options carry most of the meaning.

Structured Data & Machine Readability

weight 15 pts

Zinus México

11 / 15
Inferred

Machine-readable product data appears present and consistent with standard Shopify themes; full schema validation still pending.

Photobook México

9 / 15
Inferred

Configurable options (size, canvases, design, material, cover) appear as storefront controls rather than normalized machine-readable attributes.

AI Discovery / Retrieval Readiness

weight 15 pts

Zinus México

12 / 15
Verified public

Public browsing surfaces categories by product type and mattress size, which maps well onto natural-language buying intents.

Photobook México

9 / 15
Inferred

Product families are reachable publicly, though personalization vocabulary and a very large country selector add retrieval noise.

Merchant Policies & Trust Data

weight 10 pts

Zinus México

9 / 10
Verified public

Warranty page differentiates terms for mattresses, bases, toppers and other products; CertiPUR-US certification stated publicly.

Photobook México

6 / 10
Verified public

Footer exposes refund, privacy, terms and contact links; privacy policy recently updated. Made in Mexico, delivery stated to 38+ countries.

Inventory, Price & Variant Clarity

weight 10 pts

Zinus México

8 / 10
Verified public

Prices and variant choices are public; taxes, discounts and shipping are stated as calculated at checkout.

Photobook México

4 / 10
Verified public

Example product shows “Oferta Agotado” alongside an active “Agregar al carrito” control, and prices are stated as excluding VAT.

Cart / Checkout Agent Readiness

weight 15 pts

Zinus México

8 / 15
Needs store access

Public cart behaviour looks conventional, but agent cart handoff and UCP compliance have NOT been tested. No UCP pass is implied.

Photobook México

7 / 15
Needs store access

Personalized configuration to cart handoff is unverified. UCP / agent checkout requires active testing; nothing here implies a pass.

Measurement & AI Traffic Attribution

weight 10 pts

Zinus México

2 / 10
Needs store access

AI referral attribution cannot be assessed without GA4 / Shopify analytics access.

Photobook México

2 / 10
Needs store access

No public signal on AI traffic attribution; requires analytics access.

UCP capability layers

UCP is a fixed standardized schema. Commerce4 reports three separate, non-scored checks — base protocol readiness, Shopify enriched catalog extension readiness, and decision completeness against that schema. None of them changes either store's score.

Zinus México · UCP capability layers

Not scored · UCP 2026-04-08
  • Base UCP protocol readiness

    Expected, untestedInferred

    Conventional Shopify catalog with fixed SKUs maps cleanly onto the standardized catalog schema. Not negotiated or tested.

  • Shopify enriched catalog extension (dev.shopify.catalog)

    UnverifiedNeeds store access

    Whether the extension is negotiated, and whether gift_card, collections, requires.shipping, requires.selling_plan, checkout_url and selling_plans are populated, cannot be established from public browsing.

  • Decision completeness / standardized-schema coverage

    Coverage gapInferred

    Height and firmness read as prose rather than discrete options, so part of the buying decision is not expressible inside the standardized schema even when the protocol layer works.

dev.shopify.catalog · 2026-04-08 — extends catalog search / lookup

Product fields: gift_card, collections. Variant fields: requires.shipping, requires.selling_plan, checkout_url, selling_plans. Request filter: available (default true).

Discoverable through UCP; the residual risk is decision completeness rather than protocol fit. Transactional behaviour remains NEEDS_STORE_ACCESS until tested.

Photobook México · UCP capability layers

Not scored · UCP 2026-04-08
  • Base UCP protocol readiness

    Expected, untestedInferred

    Product families are retrievable, so base catalog search and lookup should function against the standardized schema.

  • Shopify enriched catalog extension (dev.shopify.catalog)

    UnverifiedNeeds store access

    Extension negotiation and field population are untested. checkout_url and selling_plans would be the most informative fields for a made-to-order catalog.

  • Decision completeness / standardized-schema coverage

    Coverage gapInferred

    Size, material, page count, design reference and printing instructions are chosen in a builder flow, so the configuration that defines the purchase is not represented as standardized catalog fields.

dev.shopify.catalog · 2026-04-08 — extends catalog search / lookup

Product fields: gift_card, collections. Variant fields: requires.shipping, requires.selling_plan, checkout_url, selling_plans. Request filter: available (default true).

Discoverable through UCP, but part of the purchase intent/configuration may live outside the standardized schema. Completing the buyer's intent may therefore require WebMCP or merchant-specific application state — Commerce4 does not claim this currently works; it is untested and remains NEEDS_STORE_ACCESS.

Stable vs emerging

Commerce4 measures UCP compliance against the latest stable release, pinned at 2026-04-08. Unreleased work landed on UCP main between 20–22 Aug 2026 (request constraints and payment-instrument grammar). That work is watchlist only: it is not part of a stable release and must not affect compliance assessment or scoring.

  • Request constraints (UCP main, unreleased).
  • Payment-instrument grammar (UCP main, unreleased).

Conditional checklist · Hydrogen developer preview

Detection statusNOT EVALUATED

Conditional: do not apply this checklist unless a Hydrogen developer-preview storefront is actually detected. Detection has NOT been evaluated for Zinus or Photobook, and neither store is claimed to use the Hydrogen developer preview.

  • Signed-in customer / cart session attribution reaches checkout correctly.
  • Cart-level and line-level attributes round-trip without loss.
  • Identical variants carrying different line attributes stay as separate cart lines.
  • Client-side navigation still emits page views.
  • Custom route mappings and /.well-known discovery remain reachable on the customer-facing origin.
  • Concurrent cart mutations preserve quantities and totals.
  • Logging and observability capture failures rather than swallowing them.

Why these two references?

Benchmark / positive control

Zinus México

Mexico · direct-to-consumer sleep

A relatively standard DTC Shopify catalog: conventional products, clear sizes and variants, visible shipping and warranty trust data, and a normal cart flow. It sets the upper reference line for public-web readiness.

Stress test / complexity control

Photobook México

Made in Mexico · delivers to 38+ countries

Highly configurable personalized products, size/material/design choices, international selling, VAT disclosure and far more complex product semantics. It stresses every assumption a simple variant model makes.

Together they bracket the problem space: one conventional catalog where readiness should be mostly a data-hygiene question, and one personalization catalog where the purchasable unit is a configuration rather than a fixed SKU.

Evidence model

Nothing is scored as tested unless it was tested. Protocol / UCP validation and AI traffic attribution remain pending for both merchants.

Verified public
Directly observed in publicly accessible storefront content at the time of review.
Inferred
A reasonable inference from public behaviour or platform conventions, not directly verified.
Needs store access
Requires Shopify Admin, analytics access, or an active UCP / Codex test to establish.

Zinus México

Verified public
6
Inferred
0
Needs store access
2

Photobook México

Verified public
7
Inferred
1
Needs store access
2

Top verified findings

Zinus México

zinus.mx
  1. 01StrongVerified public

    Store publicly identifies as Shopify

    Public storefront signals a standard Shopify stack, which makes Storefront API, catalog feeds and future UCP surfaces realistic next steps.

    Recommended fix
    Use the Shopify Storefront/Admin surfaces in the next layer to validate catalog completeness against what agents can retrieve.

  2. 02StrongVerified public

    Product pages expose price, identifiers, sizes and heights

    Titles, SKU-like identifiers, price, multiple mattress height and size choices, a shipping claim and detailed descriptions are all publicly visible.

    Recommended fix
    Keep this baseline and promote height/size to explicit structured attributes so agents do not have to parse copy.

  3. 03StrongVerified public

    Trust data is strong and machine-quotable

    A 10-year mattress warranty, CertiPUR-US certification, and a warranty page with differentiated terms for mattresses, bases and toppers are all public.

    Recommended fix
    Expose warranty duration per product type as a structured field so agents can answer warranty questions without reading prose.

  4. 04CriticalVerified public

    A test/demo product is publicly searchable

    “Colchón de Memory Foam | Sueño Fresco Demo pruebas” at MX$20 is reachable from public search. Agents crawling the catalog may treat it as a real, extremely cheap offer.

    Recommended fix
    Unpublish or restrict the demo product from all sales channels, search and feeds; review for other staged SKUs.

  5. 05Needs WorkVerified public

    Variant copy is less specific than the variant itself

    Marketing and product copy vary in specificity across variants, so the exact selected height is not always reflected in the description an agent reads.

    Recommended fix
    Normalize per-variant attributes (height, size, firmness) so the machine payload matches the selected variant exactly.

Pending active validation

  • · UCP / agent cart and checkout have not been actively tested.
  • · WebMCP in-session tools have not been exercised in a Chromium origin-trial session.
  • · AI traffic attribution requires analytics access.

Opportunities

  • · Review the publicly searchable demo product for indexing and catalog exposure.
  • · Normalize per-variant attributes such as exact selected height versus generic description.
  • · Expose warranty terms per product type as structured data.

Photobook México

photobookmexico.mx
  1. 01StrongVerified public

    Store publicly identifies as Shopify

    Public storefront signals a standard Shopify stack running a heavily configurable personalization catalog.

    Recommended fix
    Plan the catalog extraction layer around configurable options rather than simple variants.

  2. 02CriticalVerified public

    Stock state and add-to-cart control contradict each other

    The example Toledo Premium page presents “Oferta Agotado” while still rendering an “Agregar al carrito” control. An agent reading state machine-first cannot resolve availability.

    Recommended fix
    Make availability a single authoritative machine field and keep the cart control consistent with it.

  3. 03CriticalVerified public

    Apparent compare-at / current price inversion

    On the same public product page the current price shows MX$513 while the struck-through reference price shows MX$389. This is an apparent inconsistency observed publicly, not a confirmed backend error.

    Recommended fix
    Audit compare-at price assignment for configurable products so the discount narrative an agent reads is coherent.

  4. 04Needs WorkVerified public

    Prices explicitly exclude VAT

    The product page states prices do not include VAT, while the store sells internationally to 38+ countries.

    Recommended fix
    Publish tax-inclusive/exclusive representation explicitly per market so agents can compare landed prices correctly.

  5. 05Needs WorkVerified public

    Rich configuration is not normalized for machines

    The Toledo Premium example exposes size, “Lienzos”, “Diseño”, quantity, material, cover construction, UV texture and many size options — expressed as storefront controls rather than typed attributes.

    Recommended fix
    Model configuration as explicit machine-readable semantics: size, pages/canvases, design service, cover/material, production workflow.

Pending active validation

  • · UCP / agent checkout has not been actively tested.
  • · WebMCP configuration → LINE attributes (plus cart-wide metadata) → checkout fidelity has not been exercised in session.
  • · AI traffic attribution requires analytics access.

Opportunities

  • · Normalize configurable options into explicit machine-readable semantics.
  • · Clarify tax-inclusive versus tax-exclusive price representation for agent consumers.
  • · Resolve stock / add-to-cart contradictions and compare-at pricing inconsistencies.
  • · Reduce semantic noise from the large country/region selector in agent-readable pages.

Next validation layer

None of these steps have been executed. Each requires merchant authorization or an active test harness.

  • 1

    Storefront Catalog / UCP discovery

    Pending

    Authorized read of catalog, variants and option structures, plus discovery of any agent-commerce surface the merchant exposes.

  • 2

    UCP cart validation

    Pending

    Exercise the current UCP cart lifecycle against a test cart. Storefront MCP cart tools are being deprecated in favour of UCP cart MCP, so validation targets the UCP surface rather than the legacy tools.

  • 3

    Checkout handoff

    Pending

    Validate delegated purchase handoff end to end, including landed cost, taxes and availability truth.

  • 4

    2×2 buyer-agent evaluations

    Pending

    Run 25 UCP evals per store plus a WebMCP subset in session, recording discovery, correct product, correct variant, browser-action success, cart fidelity and checkout readiness separately.

  • 5

    WebMCP session harness (Chromium origin trial)

    Pending

    Drive the storefront's default WebMCP tools inside a real browser session and record which tools are exposed and which actually answer.

  • 6

    Line- and cart-attribute fidelity

    Pending

    Write per-item configuration as LINE attributes and verify them by reading the cart line back. The cart-attribute update/change event is observed only for cart-wide metadata (gift message, order note, provenance). Verify identical variants with different line attributes stay separate lines, and assert everything survives to checkout.

  • 7

    GA4 / Shopify AI referral attribution

    Pending

    Instrument and read AI-assistant referral segments so agent-driven sessions and revenue become measurable.

Commerce4 hypothesis

Zinus México

Zinus tests whether Commerce4 correctly recognizes a mature, conventional catalog and resists inflating its score where private testing is still missing.

Photobook México

Photobook tests whether Commerce4 can diagnose complex personalized commerce, where the product is a configuration rather than a fixed SKU.

Move from public inference to tested evidence.

Active UCP validation converts the pending categories into measured results.