Fraud center
A rule-based queue of suspicious returns. Flagged returns don't enter the fit-graph — it protects signal quality.
Blocked / recovered amount
Risk-scored queue
Sorted by score, signal, and customer history. Open the row, make the call.
False-positive trend
Rate of flagged-then-cleared, 12 weeks
There's no backend endpoint for the false-positive trend (fraud/queue only returns the instant queue).
Most frequent signals
Distribution of rules triggered in the queue
No signal in queue
No triggered rule signal in the suspicious-return queue.
Fraud rule settings
Band thresholds
Determines which band the score falls into — these two thresholds are connected to live scoring.
Rule thresholds
Default · unsavedDefault thresholds — there's no saved fraud-policy endpoint yet, these values are an unsaved draft.
| Rule | Code | Threshold | Queue hits | Status |
|---|---|---|---|---|
Serial returns | RULE_SERIAL_RETURN | returns / 90d | 0 | |
Value anomaly | RULE_VALUE_ANOMALY | × average | 0 | |
Wardrobing | RULE_WARDROBING | photo threshold | 0 | |
Address mismatch | RULE_ADDRESS_MISMATCH | mismatch | 0 | |
High frequency | RULE_HIGH_VELOCITY | returns / 48h | 0 | |
Unused label | RULE_LABEL_INTACT | label intact | 0 |
Band guide
Rule threshold exceeded. The return is set to EXCLUDE from the fit-graph. Review the case before blocking.
Instant credit was granted, the goods haven't been verified yet. Finalize the decision once received.
Low risk. It proceeds through the normal flow, and if it's a fit signal, it feeds the graph.