Methodology

How we score, what we surface, and what we will never claim

Sentinel Graph is built for legal, investigative, and compliance professionals who need traceable evidence — not black-box risk scores. This page documents our scoring model, the signals we use, the evidence quality we hold to, and the limits of what an automated platform can prove.

Tier system

Lead confidence is calibrated, not categorical

Every lead carries a numerical score (0–100) and a tier label. Thresholds reflect the strength of underlying signals and the quality of entity resolution. Customers can filter, sort, and queue by tier.

Platinum

Score ≥ 80

Highest-confidence lead. Multiple corroborating signals across independent sources, including direct exclusion linkage or active enforcement record.

Example patterns

Excluded individual confirmed as registered agent of an active provider
Subject named in DOJ press release with NPI match in active billing
OIG-LEIE record cross-referenced with property + corporate filings

Gold

Score 60 – 79

High-confidence lead. Two or more independent signals with strong identity resolution. Worth investigator triage.

Example patterns

Family-network linkage to excluded individual with shared business address
Corporate officer overlap across multiple sanctioned entities
Bankruptcy + active provider enrollment within close temporal window

Silver

Score 50 – 59

Moderate-confidence lead. One strong signal plus contextual indicators. Useful for batch review and pattern-spotting.

Example patterns

Address clustering with sanctioned entity history
Single-officer corporate web with rapid formation/dissolution pattern
Court record adjacency without direct provider linkage

Bronze

Score 45 – 49

Provisional lead. Single weak signal or unresolved entity match. Surfaced for completeness but requires significant analyst confirmation.

Example patterns

Name-only match without corroborating identifier
Common-name risk flag (further disambiguation required)
Geographic-cluster signal without direct corporate linkage

Below score 45 (the “Iron” band), leads are suppressed from customer-facing queues. They remain in our staging layer for false-positive analysis and feedback-loop calibration but are not surfaced as actionable.

Signal types

Seven signal classes, weighted by evidentiary strength

A score is not produced by a single signal. Sentinel composes evidence from multiple independent classes; lone signals — even strong ones — rarely cross the platinum threshold without corroboration.

Signal	Weight	What it surfaces
Exclusion linkage	Highest	Direct or indirect connection between an active provider/entity and an OIG-LEIE, SAM.gov, or state-list excluded individual. Includes family-name, address, registered-agent, and corporate-officer overlap.
Enforcement adjacency	High	Subject named in court records, DOJ press releases, settlement filings, or state attorney general actions. Cross-referenced against active provider enrollment.
Registered-agent webs	High	One registered agent across many sanctioned or recently-formed entities — a classic shell-network pattern. Sentinel surfaces clusters of 5+ entities sharing an agent of record.
Address clustering	Medium-High	Multiple entities sharing a single residential or non-commercial address. Mail-drop and CMRA addresses are flagged separately. Used in combination with other signals.
Corporate-officer overlap	Medium-High	Common officers across entities with sanction history, rapid formation patterns, or post-exclusion operational continuity.
Temporal anomalies	Medium	Rapid formation/dissolution cycles, post-sanction reconstitution, or NPI activity inconsistent with declared corporate status.
Asset signals	Variable	Public-record property, vessel, or aircraft ownership cross-referenced against eligibility-restricted programs (SNAP, Medicaid, low-income housing). Surfaced as eligibility gaps, not fraud determinations.

Evidence framework

Every finding is sourced, graded, and traceable

The evidence chain is the product. A score without the underlying records is not actionable for a state OIG, an SIU, or a federal investigator — so we ship both.

Source-cited

Every observation links to a specific public record with date of access. Cross-source corroboration counted explicitly.

Resolution-graded

Each entity match carries a confidence rating: identifier match (NPI, EIN), strong attribute match (full name + DOB or address), or attribute-only.

Analyst-reviewed

Findings above tier 2 pass through human analyst review before delivery. Common-name risk and false-positive suppression applied.

Lead vs. case

Sentinel distinguishes new leads (unprosecuted patterns) from proven cases (already-adjudicated subjects in our reference catalog). The two are surfaced separately and never conflated.

Limits

What Sentinel does not — and will not — claim

We do not determine fraud, liability, or regulatory violation.

Patterns are not proof. Sentinel surfaces structural concern based on public records; determinations belong to qualified legal, investigative, and regulatory professionals with subpoena power and adjudicative authority.

We do not offer bulk export at any tier.

Every lead is delivered as a per-lead, watermarked dossier with subscriber-traceable identity. CSV and XLSX bulk export are not — and will not be — features. This protects subjects from indiscriminate harvesting and protects subscribers from liability around data-resale claims.

We do not conflate prosecuted cases with new leads.

Already-adjudicated subjects appear in our reference catalog and case-study materials only. They are never repackaged as fresh leads. Customers see clear provenance for every record: new lead vs. proven case.

We do not target large publicly-traded corporations.

Sentinel focuses on local-fraud signals: small entities, regional networks, individual operators. Fortune 500 cleanup-penalty matters and federal-contractor compliance are explicitly out of scope. A blocklist suppresses these names from our lead queue.

Want to see scoring against your jurisdiction?

We can run a closed briefing on a state, MSA, or sector of your choice — with the evidence chain shown end-to-end. Initial briefings are confidential and no-cost.

Request a Briefing

Important disclaimer: SentinelGraph provides structured intelligence based on publicly available data. It does not provide legal advice, conduct investigations, or make determinations of fraud, liability, or regulatory violation. All findings are preliminary, evidence-traceable observations intended to support — not replace — qualified legal, investigative, or compliance review. Use of this service does not create an attorney-client relationship or any professional engagement unless separately agreed in writing.