Avrix grants production access against evidence that your integration handles failure paths — not
only the happy path. Run every case in sandbox, capture requestId and the response, and keep a
one-line note of how your system behaved.
Use Environments for host and key rules, and Go-live for the cutover checklist after certification passes.
Automatic certification report#
Avrix evaluates the 28 cases below from your sandbox key's own traffic. Call
GET /api/seller/v1/sandbox/certification with the sandbox key you certify, or open
Settings → Integrations → API Keys → Sandbox developer tools in the Console, which shows the
same report per key.
curl -s "https://api.avrix.io/api/seller/v1/sandbox/certification" \
-H "Authorization: Bearer $AVRIX_API_KEY"
The report covers the last 30 days of that key's sandbox requests, webhook deliveries (including test-fire receiver challenges) and fault scenarios. It is computed on each call, and it contains request and delivery references only: no keys, tokens or buyer data. Production keys receive 422 SANDBOX_KEY_REQUIRED.
Each case has one status:
| Status | Meaning |
|---|---|
pass | The traffic proves the expected behaviour, or Avrix accepted your evidence for the case. |
fail | The traffic proves the opposite, for example a mis-signed test event answered 2xx, or a 5xx retried with a new Idempotency-Key. |
not_run | No traffic exercises the case yet. reason says what to run. |
needs_partner_evidence | What Avrix can see is consistent, but the case is about behaviour inside your systems. Submit evidence for it. |
{
"data": {
"reportVersion": 1,
"generatedAt": "2026-10-12T09:30:00.000Z",
"apiKeyId": "5b2f0d1e-7c3a-4e8b-9f61-2a4c8d0e1b37",
"environment": { "mode": "isolated", "environmentId": "0c9e6f4a-1d2b-4c3e-8a7f-5b6d9e0f1a2c" },
"window": { "from": "2026-09-12T09:30:00.000Z", "to": "2026-10-12T09:30:00.000Z" },
"summary": { "pass": 4, "fail": 0, "not_run": 20, "needs_partner_evidence": 4 },
"collectionErrors": [],
"cases": [
{
"number": 18,
"name": "Invalid signature",
"group": "webhooks",
"status": "needs_partner_evidence",
"evidenceQuality": "complete",
"reason": "Your endpoint refused the mis-signed delivery. Show that it was refused for the signature, not for another reason.",
"observations": [
{
"at": "2026-10-12T08:21:00.000Z",
"deliveryId": "9e5f3c2b-7d4a-4b6f-8c8e-2a3f4d5e6b7c",
"detail": "order.fulfilled mis-signed delivery → 401"
}
],
"partnerEvidence": "none"
}
],
"notCovered": [
{
"area": "promotions",
"reason": "Answering event shares and deals, and promotion events. Rehearse them on SANDBOX-PROMO-001 (Sandbox guide, Promotions)."
}
]
},
"meta": { "requestId": "req_av1203_example" }
}
The example shows one of the 28 cases; cases always lists all of them in case-number order.
What the report does not certify. notCovered lists the areas of the Seller API the 28 cases
do not exercise, each with the reason and where to rehearse it, so a report with every case passed
is not read as covering them: promotions, finance_statements, key_management,
allocation_requests and partner_access (the example shows the first). The list is the same for
every key; rehearse these areas yourself before go-live. See
Sandbox.
How each case is decided#
- Traffic that proves the case passes it. A fulfilled
SANDBOX-ALWAYS-001order (case 7), a replay with the sameIdempotency-Key(12), IDEMPOTENCY_KEY_MISMATCH (13), a price or territory refusal (10, 11), a correctly signed delivery acknowledged (17), both secrets accepted during a rotation (22), keys returned and the refund read back (23, 24), a request after a429that waited forRetry-After(25), and a 5xx retried after a delay with the sameIdempotency-Key(26). - Traffic that contradicts the case fails it. A sold unsellable fixture (5, 6), a mis-signed or
stale delivery answered
2xx(18, 19, and then 22), a request sent beforeRetry-Afterelapsed (25), or a failed order retried under a newIdempotency-Key(14, 26). - Cases about your own systems need your evidence. Avrix cannot see a PSP refund, durable key
storage, a delisted product or why your endpoint refused a delivery. Cases 1 to 6, 8, 9, 14,
18 to 21 and 28 become
needs_partner_evidenceonce the related traffic exists; cases 15, 16 and 27 always need evidence. - Missing evidence is never a pass. When a source cannot be read, the report lists it in
collectionErrors, marks the affected casesevidenceQuality: "unknown"and does not pass them. Retry later.
Fault scenarios and test-fire produce the traffic for the hardest cases: timeout_then_success
or error_after_commit for case 14, webhook_delay_duplicate for case 21 (see
Sandbox), and the invalid_signature, stale_timestamp, duplicate and
previous_secret_only challenges of POST /webhooks/{id}/test for cases 18 to 20 and 22 (see
Webhooks).
Submitting evidence#
For a case in needs_partner_evidence, open the report in the Console and choose Add evidence:
describe how your system behaved, with request ids and timestamps, and optionally link to logs or a
recording (an https:// link). One submission per case can wait for review at a time. Avrix
reviews it: an accepted submission turns the case into pass on that key's report
(partnerEvidence: "accepted"); a rejected one leaves the case open, and you can submit again.
Evidence format#
Use this format for evidence you submit, and for any case you report outside the automatic report. For each case record:
- Case number and name
requestIdfrom the response (orX-Request-Id)- HTTP status and error
codeif any - One line describing your system's behaviour (refund issued, webhook deduped, catalog delisted, …)
A test-suite log or structured log excerpt is enough. The goal is demonstrable behaviour, not a formal document.
Required test cases#
Catalog#
| # | Case | Expected | Fixture |
|---|---|---|---|
| 1 | Cold-start sync | Full sellable catalog persisted with cursor | — |
| 2 | Incremental sync | Deltas applied, cursor advanced | — |
| 3 | Tombstone handling | Removed SKU delisted from storefront | — |
| 4 | Allocation merge | Sellable set built from GET /allocations rows with sellable: true, joined to the catalog on the SKU id | — |
| 5 | Unsellable SKU excluded | Shared-but-unallocated SKU has no buy button | SANDBOX-NOALLOC-001 |
| 6 | Non-live product excluded | Not offered for sale | SANDBOX-LIFECYCLE-001 |
Case 4 verifies you did not build the catalog from /products alone. See
Catalog and allocations.
Orders#
| # | Case | Expected | Fixture |
|---|---|---|---|
| 7 | Happy path | Order fulfilled; key delivered and persisted | SANDBOX-ALWAYS-001 |
| 8 | Preview gating | No charge when canFulfill is false | — |
| 9 | Sold out after payment | 409; buyer refunded at your PSP | SANDBOX-NOKEYS-001 |
| 10 | Price drift | Mismatch handled; no silent repricing | — |
| 11 | Territory rejection | Handled before or at checkout | — |
| 12 | Idempotent replay | Same key + same body → original response, one fulfilment | — |
| 13 | Idempotency conflict | Same key + different body → IDEMPOTENCY_KEY_MISMATCH | — |
| 14 | Unknown outcome | Timeout → replay same key; no double charge, no double pull | — |
Case 14 is the highest-value test. Simulate it by killing the connection mid-request, or with the
timeout_then_success and error_after_commit sandbox scenarios. See
Order lifecycle.
Fulfilment#
| # | Case | Expected |
|---|---|---|
| 15 | Key persistence | Key written to durable storage before responding to the buyer |
| 16 | Key non-recovery | Confirm GET /orders never returns plaintext keys |
Webhooks#
| # | Case | Expected |
|---|---|---|
| 17 | Valid signature | Accepted and processed |
| 18 | Invalid signature | Rejected — deliberately corrupt one |
| 19 | Stale timestamp | Rejected beyond 300 seconds |
| 20 | Duplicate delivery | Second delivery ignored via event id |
| 21 | Webhook before response | Handler tolerates an unrecorded order |
| 22 | Secret rotation | Both secrets accepted during overlap window |
Case 18 is the one integrations most often skip. An endpoint that accepts every signature is worse
than no verification. Send a mis-signed delivery with POST /webhooks/{id}/test and
"challenge": "invalid_signature".
Refunds#
| # | Case | Expected |
|---|---|---|
| 23 | Key return | Keys quarantined using persisted keyIds from the order response |
| 24 | Refund recorded | Refund status retrievable via GET /refunds |
Operations#
| # | Case | Expected |
|---|---|---|
| 25 | Rate limit | 429 handled with Retry-After |
| 26 | Server error | 5xx retried with backoff and the same idempotency key |
| 27 | Request correlation | requestId logged on every error |
| 28 | Reconciliation | Scheduled job compares local orders against GET /orders |
Approval#
When the automatic report shows every case as pass, or you have submitted evidence for the cases
that remain, tell your Avrix partner contact or support@avrix.io. Cases the report does not cover
for your integration can still be sent in the evidence format above.
To be told instead of checking, subscribe the key's sandbox webhook endpoint to
sandbox.certification_passed. It fires when a report for the key passes every case: the first
time, and again only if the report passes after showing a failed case. The report is evaluated when
it is built (GET /sandbox/certification, the console panel, or Avrix accepting your evidence), so
a pass that comes from new traffic is announced the next time the report is built. See the
Event reference.
Certification does not enable production access automatically, and an all-pass report does not
by itself grant it. After review, Avrix enables production Seller API access on your company. Until that happens,
production keys on api.avrix.io receive 403 PRODUCTION_ACCESS_NOT_APPROVED even after you
have signed the API Revenue Share Channel Terms and created a production key. Sandbox continues
to work for regression testing.
Avrix publishes no review turnaround. Ask your partner contact for an expected date when you submit,
and plan your cutover around it. If a case fails review, fix it and resubmit only the failed rows with fresh
requestIds.