Reference
Service level
What we commit to, how availability and response time are measured, and credits.
What we commit to, how it is measured, and what you are owed when we miss it. Your own figures are in the dashboard under Service level, month by month.
The commitment
Sandboxnone | No availability or response-time commitment. The numbers are still measured and shown to you — they are simply not something you can hold us to. |
Starter99.5% · 24h | First reply within 24 hours, 8 when urgent. 10% credited if availability is missed. |
Growth99.9% · 8h | First reply within 8 hours, 2 when urgent. 15% credited. Served first under load. |
Business99.95% · 4h | First reply within 4 hours, 1 when urgent. 25% credited. Served first under load. |
Enterpriseagreed | Written into the contract, including response times. |
How availability is measured
From your own request log — what your traffic actually did — not from a probe hitting our health check. A probe measures whether one endpoint answered one caller, which is why most status pages stay green during outages their customers can see.
- 5xx counts against us. We broke, or something we depend on did.
- 4xx does not. That is a request we refused because it was malformed or not permitted. Counting it would mean your own integration bug earned you credits.
- 429 does not. The rate limiter working is a capacity conversation, not an outage.
iNo traffic means no measurement, not 100%
How far back it goes
Request logs are kept for thirty days — they are a debugging aid and the fastest-growing table we hold. Availability is therefore rolled up daily, a few hours after each day closes, so months outlive the log they were measured from. A month is a sum of days and is kept indefinitely; the console shows the last twelve.
!A month we cannot fully account for is not a monthly figure
This mattered in both directions before the rollup existed. A month past the thirty-day horizon reported zero requests, which read as “you did not trade” rather than “we no longer hold the evidence”; and the previous month — the one that carries a credit — was computed from whatever fraction of it had survived pruning.
How response time is measured
From the moment a ticket is raised to the first reply a person can read. Internal notes we leave for each other are excluded deliberately — writing one is not answering you, and counting it would make the number look good while nobody had replied.
Credits
When a month misses the availability target, the dashboard shows the shortfall and the amount owed. Raise a ticket and we apply it.
!Credits are not applied automatically yet
What "served first under load" means
When the platform is under more load than it can serve, it sheds traffic rather than falling over — and it sheds from the bottom: unauthenticated traffic first, then Sandbox and Starter, and only past its own capacity anything above that. Shed requests get a 503 with Retry-After.
Settlement is never shed, on any plan. Wins, refunds, rollbacks and crash cash-outs always complete. Refusing a bet is survivable; refusing the win that follows one would leave a player short and a ledger half-written. Health and status are never shed either.
Your client must treat a 503 as retryable — see rate limits.