National · all states and union territories
Synthetic demo data
Overview

Trend analytics

Where the pattern breaks, and how often.

National · all states and union territories · 168 works · 150 detections

Detections per month by anomaly family

18m
Detections per month by anomaly family0255075100Oct 2024Jan 2025Apr 2025Jul 2025Oct 2025Jan 2026FinancialExecutionProcurement
FinancialExecutionProcurement

Cost per beneficiary against reach

each dot is one work
Cost per beneficiary against beneficiary count6k13k19k25k6k13k19k25kBeneficiaries reportedRupees per beneficiary
Within normFlagged (High or Critical)

Works in the upper-left region spend the most per person reached. That is not wrong on its own — a health sub-centre legitimately costs more per head than a street light — so the detector compares each work only against its own category cohort.

Detection density

total per month
Detection density by period
fewermore

Density rises through the last quarter of the financial year, when releases, bills and completion certificates are all compressed into a few weeks. Detector D-08 is calibrated for that seasonality so ordinary year-end activity does not flood the queue.

Detections by detector

alerts raised in scope
Ranked bar chartD-01 Cost overrun against sanctioned estimate26D-06 Predicted completion overrun19D-02 Unit rate above district schedule of rates18D-03 Near-duplicate work recommendation16D-05 Payment without physical progress16D-08 Fiscal year-end payment burst12D-04 Sanction splitting below approval threshold11D-11 Asset geo-tag outside sanctioned ward10D-09 Idle funds beyond permissible period8D-07 Vendor concentration in a constituency5D-10 First-digit anomaly in bill values5D-12 Work outside permissible category4

Cost per beneficiary by category

rupees, mean
Ranked bar chartSolar street lighting775Sanitation578Public library & IT560Health & wellness529Drinking water501Education infrastructure498Sports infrastructure467Passenger amenities433

Category medians are the baseline every unit-rate comparison is drawn against.

Vendor concentration

top contractors by value
ContractorAwardsDistrictsValueFlagged
Girnar Techno Civil14129.99 Cr0
Aravalli Civil Works18169.38 Cr22
Konark Buildcon16138.19 Cr25
Kaveri Infrastructure17158.14 Cr29
Gomti Builders12117.61 Cr8
Trinetra Contractors15127.49 Cr20
Shreeji Infra Works14126.25 Cr29
Neelkanth Enterprises985.88 Cr11

“Flagged” is the share of a contractor's awards sitting at High or Critical risk. Concentration alone is not misconduct; concentration together with a high flagged share is what raises D-07.

Model card

ensemble v4.2
Precision @ top 100
0.81
share of the highest-scored alerts confirmed by reviewers
Recall on labelled cases
0.74
against closed audit findings from prior years
False-positive rate
0.19
at the Critical threshold of 82
Population drift (PSI)
0.08
within the 0.10 retraining band

Scores are decision support, not findings. Every alert is a hypothesis for a human reviewer to confirm or reject, and each disposal is written back as a label — so precision improves fastest in the districts that use the system most. Detector definitions are published to reviewers in full; nothing here is a black box.