Every firm, judge, provider, insurer and asset is resolved to one record, with its source and dates, and linked to what it touches. Each product reads from the same connected record.
Illustrative. Two positions that share a firm are read as one exposure.
The methods behind the record
A connected record is only useful if every link in it can be trusted, dated and traced. These are the rules it is built on.
In short: Criterica resolves every firm, judge, provider, insurer and asset to one dated, sourced record, measures concentration across the counterparties positions share, and states a figure only where the evidence and data rights support it.
Identity: a name is never enough.
Two records join only when a key other than the name agrees: a registration number, a tax identifier, a docket. Spelling variants, common names and trade names stay separate until evidence says otherwise. A held-out set of pairs with known answers, same party or different parties, is run against the resolver on every change, and the build fails if the false-merge rate rises above its limit. We would rather miss a match than merge two parties.
Valid time: no hindsight.
Every link records when it held in the world and, separately, when we observed it. A question asked as of a date sees only the links that were valid on that date. The attorney who represented a claimant in 2021 is not treated as the claimant’s attorney today.
A fixed vocabulary of relationships.
Link types come from one reviewed list. A writer cannot invent a relationship type, and an unknown type is rejected rather than stored. Adding one is a reviewed change to the record.
Observed, asserted, inferred.
Every link states whether it was observed in a source record, asserted by a customer or inferred by a model. An inferred link can direct a review. It never becomes a fact, a lien or a finding without evidence, and no language model writes a relationship into the record.
Connected concentration.
A book can look diversified by count and still rest on a few shared counterparties. We project positions onto the firms, judges, providers and insurers they share, find the connected groups, and measure concentration twice with the Herfindahl-Hirschman index, the sum of squared shares: once with each position on its own, once with each connected group as one exposure. The gap between the two is the concentration a position-by-position view does not see.
Change travels, within bounds.
When something changes on a counterparty, the change walks a fixed number of steps along links that are valid at that moment, reaches every position it touches, and opens one task for each, with the reason, the source and an owner. The walk stops at its bound, so one event does not spread across the whole record.
Patterns are flagged, never concluded.
Three-party patterns among firms, providers and originators that recur more often than expected are flagged for a person to review. A flag is a signal for review, not a finding about anyone.
One recovery, several claims.
A single recovery can carry an advance, a provider’s lien and a firm’s fee. Each is its own payment right, allocated in priority order from the one amount that actually arrived. Allocations never add up to more than the recovery, and any shortfall is shown.
Figures only where rights and coverage allow.
A cohort figure appears only when the data rights behind every row permit the use and coverage against a stated population is known. Until both hold, the page states what is missing instead of a number.
What the record does not claim.
A shared connection is not proven correlation. Whether a link matters for an outcome is tested on history, with later information kept out of earlier tests. Calculations, allocations and policy checks are deterministic code, so the same file gives the same number every time.
- Entity resolution (name-independent)
- Joining two records only when a key other than the name agrees, so spelling variants never merge two parties and one party is never counted twice.
- Valid time
- The dates a relationship actually held in the world, recorded separately from when it was observed, so a question asked as of a date sees only what was true then.
- Observed, asserted and inferred links
- A label on every relationship stating whether it came from a source record, a customer’s statement or a model, so an inference can prompt a review but never stands as a fact.
- Connected concentration
- Concentration measured with each group of positions that share a firm, judge, provider or insurer counted as one exposure, compared with the same book measured position by position.
- Bounded propagation
- Tracing a change on one counterparty along currently valid links for a fixed number of steps, so every affected position is found without the change spreading across the whole record.
- Payment right
- One specific entitlement to cash from a recovery, such as a fee, a lien or an advance, allocated in priority order so the rights together never exceed the recovery.
- Rights-gated benchmark
- A cohort figure stated only when the data rights behind every row permit the use and coverage against a stated population is known.
The same record, read for your desk
If you finance or lend against legal assets:
Your book is read as the exposures it actually holds, not the count of positions in it. When a firm, insurer or provider changes, every position it touches is named, with an owner.
If you run a law firm:
Your matters are tied to the record as it stood on each date, so a past decision can be re-read with only what was known then. Payment rights on a matter are kept apart, so a fee, a lien and an advance never claim the same dollar twice.
If you hold medical receivables:
Each receivable is read against the rules that establish a right to be paid: authorization, contract, denial and appeal. The rule that fails is named on the row.
If you embed decisions in your product:
Each call returns one answer with its basis and is metered once. A retry returns the stored answer and is never billed again.
Structured intake captures every variable that predicts outcome.
Four-step structured intake: client profile, legal parameters, funding structure, and review. Captures jurisdiction, NOS code, parties, counsel, damages type, and procedural posture. Every field maps to a model feature.
Live scoring against the full production model fleet at submission.
Win probability, settlement probability, expected duration, P10/P25/P50/P75/P90 damages range, and risk band — computed on submission against the jurisdiction-specific model for this case type. Not an opaque confidence score — a calibrated probability tied to observed outcomes.
IC Memo auto-generated. Intelligence Panel populated.
A 9-section, bank-grade Investment Committee memo generated automatically. Executive Summary through Recommendation — legal merits, judicial intelligence, financial projections, settlement analysis. Auto-computes APPROVE / CONDITIONAL / DECLINE with full audit trail.
Automated servicing cadence activates. Risk signals tracked.
Automated multi-stage servicing cadence with structured outreach and reply matching to docket. Portfolio monitoring runs without manual intervention.
Every case scored against the jurisdiction-specific production model.
Win probability, settlement curves, damages ranges (P10–P90), risk band classification (LOW through CRITICAL), and IC Memo generation on every matter. Compare cases side by side. Underwriting queue with kanban and table views.
Health score, capital velocity, concentration risk, and at-risk capital — in one view.
Role-based dashboard with portfolio health score, capital deployment by vintage, IRR tracking across settled advances, top firm rankings by MOIC, and at-risk capital by model score band. Every metric is computed, not entered.
Five live panels. Know what needs attention before it becomes a problem.
Score deterioration, escalation queue, unresponsive matters, stale cases, and unscored intake — all live. Real-time signal feed and servicing alerts embedded in the executive dashboard.
Deep judge coverage with behavioral profiles tuned on real rulings.
Judge win rates, motion grant rates, reversal rates, citation preferences, and median time-to-decision — by case type, by opposing counsel, and by venue. Judge profile cards embedded in every case scorecard.
Facility covenants, borrowing base, UCC calendar, and disbursement tracking.
Full legal finance capital operations: loan facility management with covenant monitoring, borrowing base certificate generation, UCC filing calendar with expiration alerts, disbursement approvals, and reserve adequacy tracking.
Daily AI briefing over your portfolio. Natural language questions answered.
Morning Brief generates a daily summary across your active portfolio — new signals, settlement updates, score changes, and servicing alerts — delivered automatically each morning. Ask lets you query the portfolio in plain language.
Venue outcome differentials, opposing counsel profiles, entity intelligence.
Quantify the value of forum selection before filing. Opposing counsel clustering and win rate analysis. Entity intelligence across firms, counterparties, and related matters. Know who you're up against before the first motion.
Eight-rung cadence. Five email types. Zero manual triggers.
Automated multi-stage servicing cadence with per-case reply matching to docket and automatic cadence reset on response. The portfolio stays current without manual intervention.
Run outcome scenarios before the case reaches trial.
Simulate verdict outcomes under different case profiles, jurisdictions, and judicial assignments. Understand how changing facts shift the probability distribution before filing or before the settlement conversation begins.
30+ integrated reports. Every view of the portfolio, automatically generated.
Portfolio summary, stress test, settlement forecast, vintage performance, circuit intelligence, judge exposure, concentration analysis, duration buckets, LP quarterly, waterfall, fee yield, score vs outcome, loss development, and more — all computed from live portfolio data.
Atlas surfaces the right modules for each institutional role. Every user sees the intelligence relevant to their decisions.
30+ integrated reports. Every view of the portfolio computed automatically from live data.
Diagnostic run on your own data.
You provide a portfolio export. We run it through the relevant model suite and return a structured report. The report shows what the models see — risk concentrations, blind spots, outliers, cases the market is mispricing relative to their actual probability of resolution.
Your data team plus one analyst on our side. No integration required. No production systems touched.
Audit report: ranked findings, model scores, capital-at-risk estimates, and a prioritized action list.
Air-gapped handling applies only to services where it is verified. Atlas, our workspace platform, keeps the records and files customers put in their workspace, under the retention schedule in our Privacy Policy: uploaded source files are ordinarily deleted 30 days after an engagement closes, and delivered reports are kept up to 24 months.
Ongoing model access against live deal flow.
Audit findings are validated. You begin routing new cases through the model API as they enter your pipeline. Models score each case within seconds. Alerts surface when risk thresholds are breached or underlying conditions shift materially.
One engineering sprint. REST API with documented endpoints, webhooks for breach events, and a testing environment.
Continuous score feeds, threshold breach alerts, ranked case lists, and monthly model performance reports.
Model updates push automatically. You always run the current production build — no version management, no stale models.
Intelligence embedded into your workflow.
For enterprise and institutional deployments, Criterica Intelligence embeds directly into your underwriting, claims, or case management system. Decision flags appear natively inside your existing tools. Capital allocation recommendations route through your current approval workflows without requiring a context switch.
Joint implementation team. Criterica provides dedicated integration support and a named technical contact.
Native workflow integration, custom dashboard views, portfolio analytics layer, and a full audit trail for every decision.
Criterica is not a replacement for your team. It is the intelligence layer that makes their decisions defensible in front of a committee, a counterparty, or a regulator.
What this is not.
We do not show you a curated example dataset and call it a proof of concept. The audit runs on your actual portfolio or it does not happen.
Every score, flag, and recommendation carries an audit trail. You can see which model produced it, which features drove it, and what the confidence interval is.
Criterica Intelligence does not replace your underwriters, analysts, or claims teams. It gives them structured decision support they can defend to a committee or a counterparty.
Audit engagements are standalone. Subscriptions are annual. Integration contracts include data portability and offboarding protocols. We earn continued use by producing value.
What counts as a production model
The production model count refers to entries in the Criterica model registry that have cleared the production promotion gate. The registry is the single source of truth: every entry carries a status (production, stub, experimental, suspended, or failed), a training-data vintage, a target variable, and a jurisdiction-case-type scope.
A "model" in this count is one jurisdiction-and-case-type-and-target combination — for example, plaintiff-win probability for commercial cases in a single federal district. Criterica does not build one generalist classifier and report it as many; the count reflects narrow, jurisdiction-specific models built one at a time. Stub and gated models are tracked separately and excluded from the production figure.
How a model earns its way into production
A model is only counted as production after it proves itself against real, recorded outcomes it was never shown during training, and after it has enough real cases for the specific jurisdiction and case type to be trustworthy. Models that fall short stay in stub status and are not counted as production or offered as deployable.
A model that looks near-perfect is treated as suspect, not as success. It almost always means the answer leaked into the inputs, so the model is held back no matter how strong it looks. Any accuracy figures shown elsewhere describe performance against withheld real outcomes, and illustrate the standard, not a guarantee on any single prediction.
The court-record corpus
The corpus is the deduplicated count of real, filed records in the master index. Coverage spans United States federal and state courts, international tribunals, and regulatory enforcement bodies. Specific sources and construction are proprietary. The corpus is normalized, deduplicated across overlapping records, outcome-labeled, and time-partitioned before any model trains on it.
Every record in the production training corpus is a real filed record, with its source carried through. Because the index is appended and re-deduplicated on an ongoing basis, Criterica does not publish a running total; the current count is available to institutional counterparties under diligence.
Retraining cadence
Production models are retrained against the current master record index on a periodic basis and whenever a material data append changes the underlying distribution for a jurisdiction. Models are versioned independently; a retrain either re-promotes a model under the same standard or flags it for review. Buyers always receive the current production build through the API — there is no stale-model version management on the client side.
Market-size & duration figures
Market-size figures cited on this site (for example, the size of the US litigation-finance market, corporate legal spend, or claims-reserve exposure) are third-party estimates, not Criterica measurements. These figures are drawn from published third-party industry research; full sourcing is available to institutional counterparties on request.
Operational figures such as average case duration and any reduction in duration exposure are illustrative of the modeled effect and depend on case mix, jurisdiction, and intake process. These figures are illustrative of the modeled effect, derived from docket-level historical data; the underlying basis is available to institutional counterparties on request. They are not predictions or guarantees of any individual outcome.
Statistics shown reflect historical or illustrative model outputs derived from real case data. They are not predictions or guarantees of any individual outcome. Litigation results depend on facts, jurisdiction, judge, and counsel, and vary case by case. Model accuracy is subject to selection effects and changing legal dynamics.
Criterica Intelligence Research (2026). Methodology: How the Numbers Are Derived. Technical Note TN-001, version 1.0. Criterica. https://critericaintelligence.com/methodology
@techreport{criterica_tn001_2026,
author = {{Criterica Intelligence Research} and Cosentino, Jack},
title = {Methodology: How the Numbers Are Derived},
institution = {Criterica},
type = {Technical Note},
number = {TN-001},
year = {2026},
month = oct,
note = {Version 1.0},
url = {https://critericaintelligence.com/methodology}
}DOI: pending