Fraud & anomaly review

AI/ML anomaly engine — scores transactions against history for each card, dealer, and FPS

27
Awaiting review
6
High severity (>80)
0
Transactions blocked
99.2%
Cleared as legitimate

Review queue

Ranked by anomaly score · Nandurbar pilot · June 2026 cycle
TransactionScoreAnomaly reasonFPS / DealerStatusDecision
TXN-7af9c1
26 Jun · RC-2604...118
91
Impossible velocity
Same card lifted at FPS 142 & FPS 211, 18 km apart, within 9 min
142 / R. Sonawane In review
TXN-7af8b3
26 Jun · RC-2618...094
84
Short-weighing pattern
Load cell read 8.6 kg vs 10 kg entitlement, 14th time this dealer this month
088 / V. Patil In review
TXN-7af6d0
25 Jun · cluster
67
Exception clustering
31 biometric override approvals at FPS 211 in one hour
211 / S. More In review
TXN-7af5a2
25 Jun · RC-2641190233
22
Off-cycle lift
Sugar lifted 6 days after wheat — within normal range, low confidence
142 / R. Sonawane Cleared Auto-cleared · low score
Every decision — cleared or confirmed — is written to the audit ledger. Confirmed cases feed dealer scorecards; the beneficiary is never penalised for a dealer-side anomaly.

Active detectors

On-prem models · explainable, human-in-the-loop
Impossible velocity
Same card lifted at distant FPS within travel-impossible window
6
Short-weighing
Load-cell weight below entitlement beyond tolerance
11
Exception clustering
Unusual concentration of overrides / failures per dealer
7
Velocity on dealer
Transaction rate far above dealer's own baseline
3

Dealer scorecards

Confirmed anomalies roll up to performance, not punishment of citizens
Dealer · FPSLiftsFlagsScore
R. Sonawane · 1424,2102A · 98
V. Patil · 0883,88017D · 61
S. More · 2112,9409C · 78
D. Gavit · 3053,1101A · 96
Scorecards are a Horizon-2 capability, previewed here on pilot data.