Common questions from institutional buyers.

Most legal data providers are document stores. They give you access to case filings, dockets, and opinions — raw text at scale. That is useful for research. It is not useful for prediction.

Criterica's 475M+ court record dataset is structured for outcome modeling. Every record is normalized, deduplicated, enriched with outcome labels, and mapped to a jurisdiction-specific schema. Our proprietary outcomes corpus covers all federal district and appellate courts from 1960 forward. The dataset includes outcome labels (plaintiff win, settlement, dismissal), case duration, damages awarded, judge assignment, circuit, district, case type, and financial party data where available.

The distinction matters because raw documents cannot be fed into a classifier directly. Structuring legal data for predictive modeling requires jurisdiction-specific label logic, deduplication across overlapping sources, and careful handling of partial records. That pipeline took years to build and is not available from any existing data vendor.

Criterica Intelligence runs 23,706 production models. Coverage spans:

Federal: All 12 federal circuits and 94 district courts across all major case types (civil rights, commercial, IP, securities, antitrust, mass tort, employment, healthcare, bankruptcy, and more).

State: 30 U.S. states with sufficient volume for litigation finance and outcome prediction models. Remaining 20 states are in stub status pending volume thresholds.

International: United Kingdom (England & Wales, Northern Ireland, Scotland), Australia, Canada (BC, AB, ON, QC). International models currently cover commercial and appellate outcomes.

Stub models, those that have not yet proven how reliably they predict on cases they never saw, are visible in the platform but not deployed. They represent data-gated capacity waiting on licensed or volume-sufficient datasets.

Phase 1 (Audit): One analyst on your side for data export. We accept CSV, Excel, or direct database exports in any standard schema. The export format does not need to match our schema — we map it. Timeline is 1–2 weeks.

Phase 2 (Subscribe): One engineering sprint to integrate the REST API. Estimated 1–3 days of backend engineering. We provide a full API reference, test environment, and sample request/response pairs for every endpoint. Webhooks are available for real-time breach alerts.

Phase 3 (Integrate): Varies by platform. We have pre-built connectors for several major underwriting and case management platforms. Custom integrations involve a joint implementation team and typically complete in 6–10 weeks.

There are no infrastructure requirements on your side. Criterica Intelligence runs as a hosted API. You do not manage models, training data, or compute.

Every prediction comes with an audit trail that identifies the top contributing features, the model version that produced the score, the training cohort, and the confidence interval.

The models are tuned gradient-boosted ensembles and logistic regression variants, not deep neural networks. This is an explicit choice. Institutional buyers need to explain their decisions to investment committees, regulators, and counterparties. A model that returns "the neural network said so" is not deployable in that context.

What you can show a counterparty: the historical base rate for this case type in this jurisdiction, the specific features that pushed the score above or below the base rate, the model's performance on the validation cohort, and the audit trail for the specific case.

What the model does not claim: certainty. Every output is a probability with a confidence interval. The goal is to make uncertainty legible and structured, not to eliminate it.

Audit engagements run on an air-gapped basis. Your portfolio data is used only to run the diagnostic. It is not retained, used for model training, shared across clients, or used for any purpose other than producing your audit report.

For ongoing subscriptions, case data submitted to the scoring API is processed and returned. It is not stored in a shared environment or used to update production models.

Criterica Intelligence does not train on client data without an explicit data licensing agreement, which requires separate negotiation and a separate data governance framework.

All data transmission uses TLS 1.3. Retention policies, data processing agreements, and GDPR/CCPA compliance documentation are available upon request during the audit engagement.

A production model has proven how reliably it predicts on a cohort of cases it never saw, drawn from real adjudication data. It has also passed a minimum row threshold (500+ training rows for the specific jurisdiction-case-type combination).

A stub model has been scoped, meaning the training pipeline, feature schema, and target definition exist, but one of two gates has not been cleared: either the available training data is below the minimum row threshold, or the model did not predict reliably enough on initial training.

Stubs represent data-gated capacity. For example, several workers' compensation and medical malpractice models are in stub status because the licensed industry datasets required to train them are not yet in place. When those licenses are in place, the training pipeline runs automatically and the models either promote to production or are flagged for review.

The distinction matters for buyers: you know exactly what you are getting and why certain jurisdictions or case types are not yet available.

Pricing has three components:

Audit fee: Fixed-fee engagement based on portfolio size and audit type. Covers the diagnostic run, the findings report, and one follow-up review session. Audit fees are standalone — no commitment to a subscription is required.

Subscription: Annual contract priced by volume tier (cases scored per month) and model access level (specific jurisdiction-case-type combinations versus full fleet access). Enterprise pricing is available for multi-market deployments.

Integration: Quoted per engagement based on scope. Includes implementation support, custom connectors, and a dedicated technical contact.

API access for platforms and technology partners is priced separately and includes both volume-based and per-model-access tiers.

Specific pricing is discussed during the initial audit scoping call. We do not publish rate cards — every engagement starts with a scope conversation.