AAVAE™
AAVAE: agentic analysis after the deterministic rules
AAVAE is the Adversarial, Adjudication, Validation, and Attestation Engine, formerly AVA. It starts where the deterministic rules stop. The rules decide what is flagged. AAVAE groups the flags into cases, looks for the patterns and anomalies across them, argues the defense's side, and seals the record for a person to review.
People make every finding, every damages decision, and every decision to act. AAVAE prepares the file and refuses to seal one that does not verify.
Rules first
The rules decide what is flagged. AAVAE works on what they found.
Tier 1 is deterministic: the platform's 332 checks run the same way every time, and every pass, fail, and review trigger is recorded. AAVAE does not replace them and never runs ahead of them.
Tier 2 is where AAVAE starts. Thousands of separate flags rarely mean thousands of separate problems. AAVAE reads them against the claim history, resolves the relationships between providers and claims, and finds scheme signatures such as systematic upcoding, unbundling, and duplicate billing patterns. A reviewer receives a small set of cases with the evidence attached, not a queue of alerts.
AAVAE also challenges the rules. When it disputes a rule's pass or fail it states the reason, and the challenge goes to a reviewer. A challenge never silently overturns a rule.
Six agents
Six agents, one fixed order
Each agent has one job, a locked system prompt whose SHA-256 hash is recorded, and an output contract the pipeline checks. Each result is hashed and anchored on CourtChain™ before the next agent starts.
Investigator
Sweeps the vendor's claim history against the Tier 1 signals to set base rates and find scheme clusters. Returns the pattern signatures, a summary of the flagged cohort, and references to the evidence it used.
Theorist
Maps each pattern to the statutes and policies that could apply, such as the False Claims Act, the Anti-Kickback Statute, and the Stark Law. Gives citations, a strength for each element, and the gaps still open.
Adversary
Argues the defense. Anticipates what opposing counsel, an auditor, or a materiality challenge would say, rates how well the case answers each, and names the weakest link.
Damages Quantifier
Computes exposure by four methods: unit of overpayment, extrapolation, benefit of the bargain, and disgorgement. Amounts are integer cents with a 95% confidence interval.
Drafter
Turns the structured findings into the investigative memo, an exhibit schedule, and the bundle for counsel. Exhibits are listed from the agent runs that produced them, not estimated.
Auditor
Checks the whole chain before anything is sealed: order, hashes, output contracts, and evidentiary gaps. Returns yes or hold.
The hard gate
A record that does not verify is not sealed
The Auditor reports and the pipeline decides. Before the Auditor runs, the pipeline recomputes every hash and checks every output against its contract. After the Auditor returns, the pipeline takes the seal decision again from the same checks. An Auditor that says yes over a broken chain is overruled, and a missing decision counts as a hold.
A held case is not sealed, and no final bundle is assembled. If the hold is an evidentiary gap, a named reviewer can resolve it, and the resolution is recorded on the same anchored record. If the chain itself does not verify, nothing a person writes releases it. The case is run again.
AAVAE does not decide fraud, set final damages, or stop a payment. Reviewers, counsel, and regulators make every substantive decision.
The record
Built so someone else can check the work
Every agent runs at temperature 0 against its hashed prompt. Each result carries a ModelCard: the model identifier, the prompt hash, the region it ran in, usage counts where the provider reports them, and the time. With the same model version, the same prompt, and the same tokenized input, an expert can replay a stage and compare it. The record names the model version, so a replay depends on that version still being available.
Each stage's hash is chained to the one before it and anchored on CourtChain. The bundle is designed to support self-authentication under FRE 902(14) with a written certification. A court decides what it admits.
Your data
It runs where your data already is
AAVAE is deployed inside the customer's own environment, and customer data does not go to a public model service. Identifiers are replaced with format-preserving surrogates at the ingest boundary and the agents work on the surrogates. Re-identification happens only at the customer's authorized output.
Model, region, cost ceiling, and business associate agreement status are set per customer. If a required agreement or region setting is missing, AAVAE refuses to run. The models are used for inference only. Customer data is not used to train or fine-tune them.
Evidence bundles
What AAVAE builds for each bundle
AAVAE is the analysis behind JIL's four evidence bundles. Each bundle carries the same replayable record and the same CourtChain anchor.
Court Ready Evidence Bundle
For whistleblower counsel, state Medicaid fraud control units, special investigations units, and prosecutors. AAVAE finds the cohort, maps it to the statutory elements, tests it against the likely motions to dismiss, quantifies the exposure, and seals the record. CREB product page
Audit Ready Evidence Bundle
For health systems, provider groups, and revenue cycle teams. AAVAE runs ahead of an audit, ties billed codes to the clinical documentation, and simulates the payer's or CMS auditor's rebuttal so unsupported diagnoses surface early. Internal audit
Payer Integrity Evidence Bundle
For health plans, Medicare Advantage organizations, managed care organizations, and unified program integrity contractors. AAVAE compresses thousands of edit flags into provider-level cases, separates systemic patterns from isolated clerical errors, and produces an overpayment schedule with confidence intervals.
Risk and Recovery Evidence Bundle
For Medicare Advantage plans, accountable care organizations, and subrogation teams. AAVAE scores risk adjustment populations continuously, finds unsupported HCC submissions before an audit sample is drawn, and packages the reconciliation for recovery outside litigation. Continuous RADV readiness
See it on a case file
We walk through the architecture, the deployment options, and the agreement structure under NDA.
Request a briefing Read the architectureAAVAE, CourtChain, CREB, PIEB, and RREB are trademarks of JIL Sovereign Holdings, LLC, licensed to JIL Sovereign Technologies Incorporated. Marks are filed and not yet registered.