Catch payment anomalies before federal monitoring does.

Federal improper payments keep rising. Grant teams still review by hand.

Tap a bar for year-by-year detail

Source: GAO · OMB IPERIA annual reports

Three layers. Nine engines.

Ordered the way grant and finance teams actually work: close the gaps monitors already name, catch anomaly patterns rules miss, then document every flag for HUD review and single audit.

Layer 1 · Close obvious gaps

The improper-payment categories finance teams and auditors already expect you to catch.

NLP

Documentation Gap

Scan claims and invoices for missing signatures, wrong form types, and field mismatches. The “boring” category behind a large share of improper payments — and fully automatable.

$70B/yr exposure
SSA DMF

Deceased Recipient

Match every payee against the 90M-record SSA Death Master File. Flag payments after date of death, including overlap windows legacy batch jobs miss.

$18B/yr exposure
Document AI

Contract Intelligence

Ingest vendor contract PDFs, extract not-to-exceed and unit rates, then reconcile actual payments. Surfaces overages and invoices split just under competitive-bid ceilings.

Procurement integrity

Layer 2 · Catch what rules miss

Ensemble models and graph analysis for related vendors, spend drift, and burst payment patterns — not static thresholds.

MCMC · GNN · CNN

Ensemble Anomaly Scoring

Fuse geometric MCMC, peer GNN comparison, and payment-timing CNN signals with Z-score and Benford checks. One score for unusual spend shapes, timing, and multi-signal anomalies.

Peer-benchmarked
GNN · Entity

Vendor Clustering

Link vendors with near-duplicate names, synchronized payment timing, or identical spend fingerprints. Surfaces related-entity patterns that stay under line-by-line review.

Related-vendor detection
6-Quarter

Behavioral Drift

Segment each vendor’s history by fiscal quarter and score the trajectory. Surfaces vendors sliding from clean billing toward threshold-gaming before totals look wrong.

Early-warning signal

Layer 3 · Defend every flag

Plain-English review memos and audit-ready packets so grant staff and federal reviewers get answers, not black-box scores.

Budget Forcing

AI Case Review

A frontier model writes a review memo per flag, with a second reasoning pass on borderline scores. Staff get narrative, math, and a recommended next step.

Explainable by design
HUD · Single Audit

Audit & Compliance Readiness

Auto-assemble reporting for grant monitoring, IPERIA, Single Audit Act, and internal control reviews — tied back to each flagged payment and its memo.

Monitoring-ready
Cross-batch

Duplicate Payment Detection

Match invoices and disbursements across files and fiscal periods for same vendor, amount, and date fingerprints. Surfaces duplicates grant and AP exports miss when systems stay siloed.

Cross-file matching

From CSV to findings.

Grant administrators shouldn't need a six-month data project before screening starts. Send a vendor payment file the way your system already exports it — the normalizer handles the rest.

01

Send your CSV

Email a vendor payment export from your finance or grant system — any format. No API, no integration project, no six-month onboarding.

02

Columns map automatically

SPECTR's schema normalizer recognizes vendor, amount, date, and program fields on its own, whatever they're named in your file.

03

Screening runs overnight

Nine engines score every payment for exclusions, duplicates, threshold clustering, and peer anomalies. Each flag gets a review memo with the math behind it.

04

Report arrives by email

A plain-English findings report: what was flagged, why, estimated exposure, and recommended next steps. Free for pilot agencies.

Not an anomaly score. An argument you can audit.

Grant reviewers and federal monitors need explainability — not just a score. Most tools can't show their work. That's why teams stay with spreadsheets and manual sampling even when disbursement volume outpaces staff.

SPECTR produces a mathematically auditable probability statement, then writes it up in plain English. Here's a sample memo. If you can read it and see why the payment was flagged, the monitoring conversation is already half done.

Case Memo · Auto-generated
Flagged
Meridian Facilities Group LLC
Vendor #MFG-4471 · Building Services · 14 invoices · FY2025
0.94
ensemble score
Critical · 96% conf.
Detection typeInvoice SplittingBid-Threshold EvasionProcurement BypassStructured Payments

This vendor's spend allocation sits at the 2nd percentile of the feasible distribution for building-services vendors in this jurisdiction. Eleven of its fourteen invoices land between $24,400 and $24,900, clustered immediately beneath the $25,000 competitive-bid threshold, a pattern consistent with invoice-splitting to avoid procurement review.

Pattern
Structured invoice split
Peer group
47 like vendors
Period
FY2025 · 9 months
Threshold rule
Bid limit $25,000
2nd %ile
Geometric MCMC
allocation vector
Spend shape in bottom 2% of normal
+3.8σ
Peer GNN
vs. category peers
Far outside 47 similar vendors
χ² 41.2
Benford
leading-digit fail
Invoice amounts look hand-picked
11 / 14
Threshold clustering
just under $25k
Invoices bunched under bid limit
Recommended action: route to procurement review · request unsplit work orders · est. exposure $341K
2nd reasoning pass · budget-forced

Built on sources any agency can access.

Federal grant programs cross-reference public exclusion lists on day one. State and local grant offices administering CDBG, disaster recovery, and infrastructure funds are where proactive screening is thinnest — and where we start pilots.

SSA Death Master File

90M+ deceased beneficiary records

CMS NPI Registry

Every licensed US provider

USASpending.gov

All federal payment transactions

IRS income thresholds

Published AGI eligibility bands

SAM.gov

Debarred & excluded contractors

OIG LEIE

Excluded parties list

State APCD

All-payer claims where published

State DOC records

Incarceration status where shared

Public data is the start. Agency feeds unlock the rest.

Standard state and county procurement opens MMIS exports, eligibility rosters, and vendor payment files under existing data-use agreements. SPECTR layers the same cross-reference intelligence federal agencies pay tens of millions to configure, scoped for state budgets and 90-day pilots, not multi-year sole-source awards.

01
State MMIS feeds

Direct Medicaid claims ingestion, the data state HHS already exports daily.

02
County vendor payments

Contractor invoice files from procurement systems, cross-referenced against SAM.gov in real time.

03
Eligibility & APCD files

Member eligibility rosters and all-payer claims history under standard data-use agreements.

04
Cross-agency matching

The same duplicate and deceased-recipient detection federal agencies pay tens of millions to configure.

Free pilot

Send us your vendor payment CSV. We run the analysis for free. You get the report.

No procurement cycle to start. No integration project. Email us a vendor payment file, columns map automatically, and we send back a plain-English findings report with every flag explained and the math shown. If it's useful, we talk about a pilot. Grant compliance and audit-ready applications are Navigator — request a demo for that.

CSV stays yours · analysis runs on your file only · no data retained after the report