Readiness scoring

A score is not the product. Knowing which stage broke, and why, is.

Nine stages, scored the same way every time, with a stable cause code and a concrete fix behind every failure.

Stages: 9
Scoring: readiness-v2
Cause codes: 41 · taxonomy-v1
Merge rule: worst agent style wins

The need

“Agent-ready” is not a yes or no. It is nine separate questions.

An agent can find your product, read its price, and still be unable to buy it. Another can check out perfectly and fail at discovery because your search returns a page it cannot parse. A single readiness number hides which of those you have.

So tragent scores the journey in stages — discovery, product data, price, inventory, cart, checkout, payment, refund, authentication — and reports each one separately, with the evidence that produced it.

SCHEMA_OFFER_MISSING
Name and image are marked up. Price, currency and availability are not, so the agent guesses or moves on.
BOT_RULE_BLOCKS_AGENT
A broad bot rule returns 403 to declared commerce agents alongside the scrapers it was written for.
UCP_PROFILE_INVALID
The /.well-known/ucp profile advertises endpoints that 404 or capabilities the API never implemented.

A crawler check tells you a page is reachable. It cannot tell you whether an agent that reached it could complete a purchase.

How it works

Two kinds of evidence, scored by one algorithm.

Deterministic checks
schema.org Product, Offer, price and availability extraction; /.well-known/ucp validation; robots and agent-policy analysis; bot-block requests sent with declared agent user agents; transport security. Fast, cheap, and identical on every run.
Judged journeys
What the simulated agents actually managed, stage by stage, with the step that produced each verdict attached as evidence.
Weighted, then merged
Stages carry weights — discovery and checkout 3, price, inventory, cart, payment and authentication 2, refund 1. Observations average within an agent style, then the worst style wins across them.
Skips are excluded, not zeroed
A stage the run never attempted is recorded as skip and left out of the mean. A search-only test does not make your checkout look broken.

StagesTest plan retail-core · 2 agent styles

passFind productsfound 3 matching products via searchPassed
passRead the detailsschema.org Product + Offer presentPassed
passSee the price$89.00 USD matches visible pricePassed
passCheck stockInStock · size 10 availablePassed
passAdd to basket1 item · subtotal $89.00Passed
warningReach checkoutauthorisation header rejected on 2 of 5 triesWarning
failPaydelegated payment not acceptedFailed
failReturn an orderreturns need a visual CAPTCHAFailed
warningSign inno delegated sign-in availableWarning

Every row carries its own evidence. The status is the headline; the line beside it is the reason.

What it brings

Findings you can hand to an engineer without translation.

A stable code for every failure
41 cause codes in a versioned catalogue, each with a title, a severity and a concrete fix. The same defect gets the same code next quarter, so you can count it.
Scores that mean the same thing twice
The algorithm is versioned as readiness-v2 and stamped on every run. Two runs are comparable when the version matches — and you can see when it does not.
The worst case, not the average
If checkout works for one agent style and fails for another, the report says failed. Averaging that away would hide the customers you are actually losing.
Money, where you have the figures
Supply traffic, agent share, conversion and order value and each failing stage carries what it blocks — as a monthly figure, not a grade.

How the money figure is built

The revenue model, econ-v1, walks the funnel once and consumes a shared pool, so two broken stages cannot each be charged for the whole funnel. Refund scores zero because it cannot block a purchase that already happened; authentication carries half weight because guest checkout is common; a warning counts 0.35 rather than as a hard block.

econ-v1 · figures you supply, labelled as estimates

Boundaries

What it does not do.

Stated here so you do not have to discover it in a pilot.

Scores are comparable within a version, not across them.
readiness-v2 does not produce the same number as readiness-v1. The version travels with the run so a chart cannot silently mix them.
A free public scan does not test checkout.
It runs the deterministic pack against one URL. Multi-style journeys through cart and payment need a verified domain and an account.
It finds what it can reach.
Flows behind a login you have not shared, or a payment sandbox tragent has no credentials for, come back as skip — visibly not tested, rather than quietly assumed fine.

See it against your own site. A scan takes about ninety seconds and needs no account.