Pattern Lab
Fraud signal coverage
Use this page to defend the detector: per-card anomaly scoring, cross-card aggregation, explainable reasons, and cost-aware tuning.
Rows processed
1,000
The command center expects all 1,000 challenge transactions to be ingested before review.
Flagged for review
75
8% review rate at threshold 49.
High-pressure flags
21
Critical and high levels are the first pass for a seven-minute demo.
Active patterns
6
Model mimir-fraud-hil-v1 running in balanced mode.
Primary fraud patterns
Patterns are ranked by flagged volume and link back into the transaction reviewer.
Risk distribution
The distribution makes over-flagging visible before export.
critical
7
high
14
medium
75
low
904
Detector signal mix
Every flag should be explainable by at least one human-readable signal.
Model consensus
500 transactions above signal threshold
50
Category surprise
94 transactions above signal threshold
28
Card baseline
16 transactions above signal threshold
11
Shared entity graph
0 transactions above signal threshold
7
Velocity
39 transactions above signal threshold
4
Cross-card clusters
Shared devices, IPs, and merchants expose patterns a single-card baseline cannot see.