FIDELITY RECORD

Verified against real Stripe: every probe we run, and its result.

Every behavior Worlds claims to match is backed by a probe: a short script that sends the same requests to a world and to real Stripe test mode, then compares the responses. A deleted customer returns 200 with a stub. An over refund is refused with Stripe's exact wording. A replay under the same idempotency key returns the first response. The full battery runs in every test run of the product. Against real Stripe it runs whenever a test key is present, and the newest run is what you see here.

LAST RUN

2026-09-14

PROBES

20

AS EXPECTED

20 OF 20

ALL OF THEM

MUST AGREE

16

DOCUMENTED GAPS

4

PROBEEXPECTEDRESULT
customer-crud-and-deleted-stubMUST AGREEAS EXPECTED
unknown-id-404-envelopeMUST AGREEAS EXPECTED
pi-confirm-visaMUST AGREEAS EXPECTED
pi-confirm-declined-persists-failureMUST AGREEAS EXPECTED
over-refund-messageMUST AGREEAS EXPECTED
charge-already-refundedMUST AGREEAS EXPECTED
refund-listed-on-chargeMUST AGREEAS EXPECTED
idempotency-replayMUST AGREEAS EXPECTED
idempotency-param-mismatchMUST AGREEAS EXPECTED
idempotency-validation-not-reservedMUST AGREEAS EXPECTED
idempotency-validation-amount-zeroMUST AGREEAS EXPECTED
pagination-has-moreMUST AGREEAS EXPECTED
invoice-lifecycle-draft-open-voidMUST AGREEAS EXPECTED
customer-balance-credit-appliesMUST AGREEAS EXPECTED
invoice-pay-autochargeDOCUMENTED GAPAS EXPECTED
pi-capture-manual-refusedDOCUMENTED GAPAS EXPECTED
price-trial-period-daysDOCUMENTED GAPAS EXPECTED
balance-txn-invalid-currencyMUST AGREEAS EXPECTED
balance-txn-second-currency-refusedDOCUMENTED GAPAS EXPECTED
credit-other-currency-not-appliedMUST AGREEAS EXPECTED

Run 2026-09-14T19:28:39Z against real Stripe (test mode). Source: 2026-09-14.json in the product repository. Probes create only their own tagged objects in the test account and touch nothing else.

WHAT A GAP IS

A gap is a difference we keep on purpose and document: a parameter real Stripe accepts that Worlds refuses with a 400 that names it, so an agent relying on it finds out immediately instead of being graded on a guess. The battery fails if a gap closes without the docs saying so, just as it fails when an agreement drifts. The scope page lists what Worlds implements and what it deliberately leaves out.

Every live run so far has found and fixed at least one entry the same day. That is the point of the loop: the record gets more accurate with every run, and a published record of live checks cannot be copied.