What is the baseline rate calibrated against?

Each program's officially published improper-payment rate is normalized against a fixed reference rate representing a typical program-wide average, producing a baseline factor. A program with a rate at the reference value produces a neutral factor of 1.0; a program with a materially higher published rate produces a higher factor, and vice versa, within fixed floor and ceiling bounds.

Does the baseline factor change the deterministic fraud-detection scoring itself?

No. The underlying detection logic - the linkage-agreement matrix, premise-capacity checks, and similar deterministic rules - is unaffected by the baseline factor. The factor and its full source lineage (program code, agency, fiscal year, published rate, source document) are captured into the audit trail so downstream narrative reporting and evidentiary submissions can cite the exact official rate in effect at the time of detection.

What happens when no published baseline exists for a program?

The system falls back to a neutral factor of 1.0 and records that no baseline was found, rather than guessing at a rate or silently omitting the calibration step - the absence of a baseline is itself a recorded, auditable fact.

How current is the published-rate data?

Baseline rates are refreshed from the reference table on a daily cache cycle so that a newly seeded rate - published when an agency issues its annual improper-payment report - is picked up without a service restart.

Agency-Published-Baseline-Calibrated Fraud Confidence Factor

Executive Summary

Federal benefit and payment programs vary enormously in their historical improper-payment rates - a program the relevant agency has publicly reported at a low single-digit improper-payment rate carries a very different fraud-prior than one an agency has reported at a rate several times higher. JIL Sovereign's baseline-calibration mechanism resolves each engagement's primary program code to the most recently published improper-payment rate for that exact program, normalizes it against a fixed reference rate to produce a baseline factor bounded within a fixed floor and ceiling, and captures the full lineage of that calculation - program code, issuing agency, fiscal year, published rate, and the source document itself - into the engagement's audit log.

The calibration is deliberately kept separate from the deterministic detection logic itself: it does not change which patterns get flagged, only how the resulting evidentiary narrative and downstream confidence reporting can be framed against the program's own official track record, with a citable source document behind every number.

Core Innovation: Every fraud-confidence narrative is traceable to the specific agency-published improper-payment-rate document that was in effect for that program at the time of detection - not a generic industry assumption - and that provenance is written into the same evidentiary chain as the detection findings themselves.

Problem Statement

Improper payments have been a GAO government-wide High-Risk Area for more than two decades, and individual agencies are required to estimate and publicly report an improper-payment rate for their major programs under the Payment Integrity Information Act framework. Those published rates vary widely by program and by fiscal year, yet most fraud-detection tooling either ignores this context entirely or applies a single generic industry assumption regardless of which specific program a claim belongs to - which understates confidence for a program with a documented high improper-payment history and overstates it for a program with a documented low one.

Just as importantly, when a fraud finding is later challenged - in an appeal, an audit, or litigation - a confidence assessment that cannot point to the specific published source document behind its calibration is far weaker evidence than one that can.

20+ yrs

Improper payments a GAO government-wide High-Risk Area

$236B+

GAO-estimated federal improper payments, FY2023

Why Generic Priors Fall Short

One-size-fits-all fraud-rate assumptions: ignore that different federal programs have very different documented improper-payment histories.
Undocumented internal priors: even where a vendor does calibrate per-program, the source of that calibration is often internal and not independently verifiable against a public document.
Stale calibration: without a refresh mechanism tied to the agency's own publication cycle, a program's calibration can silently go years out of date.

Technical Architecture

Baseline Resolution and Factor Computation

Step	Operation
1	Resolve the engagement's primary program code
2	Look up the most recently published improper-payment rate on file for that program code (highest fiscal year available)
3	Normalize: `baseline_factor = published_rate / reference_rate`, clamped to a fixed floor and ceiling
4	Return the factor together with the program code, agency, fiscal year, and source document reference
5	Record the full tuple into the engagement's audit log at detection time

The reference rate represents a normalizing constant against which every program's published rate is compared - a program whose officially reported rate equals the reference produces a neutral factor of 1.0, a program with a materially higher reported rate produces a proportionally higher factor up to the ceiling, and a program with a materially lower reported rate produces a proportionally lower factor down to the floor. Bounding the factor prevents a single outlier program-year from producing an implausible multiplier.

Evidentiary Capture, Not Score Manipulation

The baseline factor is computed once per engagement and logged - it does not alter the deterministic linkage-agreement, premise-capacity, or other rule-based scoring described in the companion FWEA claims. Its role is confined to the narrative and evidentiary layer: every downstream report, appeal response, or litigation-support export can cite the exact program code, agency, fiscal year, published rate, and source document that calibrated the confidence framing for that specific engagement, rather than an unattributed industry assumption. Baseline data is cached in-process with a 24-hour refresh so a newly published agency rate is picked up promptly without requiring a service restart, and an engagement with no matching program code or no baseline on file explicitly records a neutral, unaugmented factor rather than defaulting silently.

Prior Art Differentiation

Approach	Program-Specific Calibration?	Source Document Citable?	Refresh Cycle
Generic industry fraud-rate assumption	No	No	N/A
Internally-derived per-vendor priors	Sometimes	Rarely independently verifiable	Vendor-discretionary
JIL baseline-calibrated confidence factor	Yes - by program code and fiscal year	Yes - specific agency document captured	24-hour cache, tied to agency publication cycle

Patent Claim

Independent Claim 120: A computer-implemented method for calibrating fraud-detection confidence reporting comprising: resolving, for a payment-integrity engagement associated with a specific government program, an officially published improper-payment rate for that program corresponding to the most recent fiscal year on file; computing a baseline confidence factor by normalizing the published rate against a fixed reference rate and bounding the result within a predetermined floor and ceiling; associating the baseline confidence factor with the program's issuing agency, fiscal year, and a reference to the source document in which the rate was published; recording the program code, published rate, fiscal year, source-document reference, and computed factor into an evidentiary audit log at the time of detection, separately from and without altering the deterministic fraud-detection scoring applied to individual transactions in the engagement; and, when no published rate is on file for the program, recording a neutral factor together with an explicit indication that no baseline was found.