FAQ
Every answer here is checked against the code that actually runs, not written from memory.
What's the difference between DEGRADED and UNTRUSTED?
UNTRUSTED means the drift alarm fired, or more than 20% of the batch was flagged — the recommended action is to block or escalate to a human. DEGRADED means the flagged rate is above twice the expected false-positive rate, but no alarm has fired yet — worth a look before it escalates, but not necessarily a reason to stop serving predictions.
What happens if I call /fit again with different data?
It replaces the trained baseline for that model. The baseline is the only ground truth the engine ever assumes — refitting on drifted or biased data will make future verdicts reflect that new baseline, not the original one. There's no versioning of baselines yet; if you need to compare against an older one, keep the training rows yourself.
Is my data used to train anything shared with other customers?
No. Each tenant's baseline, evidence, and decision history are isolated to that tenant. Nothing you send is used to improve a model for anyone else.
What happens when I hit my monthly prediction limit?
New calls to /assess are refused with a 429 response until the counter resets next month, or you upgrade. It fails closed with a clear error, not silently or with degraded accuracy.
How long is my history kept?
7 days on Free, 30 on Pro, 90 on Business, and effectively unlimited on Enterprise — see pricing for the exact numbers per plan.
Can I delete my account and everything in it?
Yes, from Account & keys. It removes the account, its models, its keys, and its decision history — checked by an automated test after every deletion, not a manual checklist someone might forget a table in.
Do you offer a free trial of the paid plans?
The Free plan itself has no trial clock or card requirement — it's free indefinitely, not a 14-day countdown. If you outgrow its limits, upgrading is immediate and self-serve.
What if the evidence behind a decision looks wrong?
Open it — every decision's evidence panel shows exactly which inputs moved and by how much, plus the raw martingale value against the alarm threshold. If the underlying data was the problem (a bad baseline, a one-off bad batch), that's visible in the evidence itself, not hidden behind a single "flagged" label.
Do you have an SLA?
Business plan includes a 24-hour support response target. There is no uptime SLA with service credits yet — see the honest state of that on the Trust & Security page rather than a number we can't yet back with real operating history.
Question not answered here? Contact us.