Golden test packs for five regulated verticals.
Browse expert-labelled suites for healthcare, finance, law, insurance and life sciences, and see which regulations each one answers to.
Browse by vertical.
Each vertical has its own reviewers, rubrics and regulatory map, and runs on the same assertion engine.
- Healthcare & ClinicalClinical documentation, prior authorization, triage and medical coding agents.4 packs · 8,630 casesBoard-certified physicians, nurses, certified coders
- FinTech & BankingUnderwriting, wealth advisory, KYC/AML and customer-service banking agents.4 packs · 6,900 casesFair-lending attorneys, registered principals, former examiners
- LegalTech & ContractsContract review, statutory research and litigation support agents.3 packs · 4,730 casesLitigation and transactional attorneys
- Insurance & ClaimsFNOL intake, coverage interpretation and claims adjudication agents.3 packs · 4,200 casesAdjusters, coverage attorneys, actuaries
- Life SciencesPharmacovigilance, clinical trial protocol and regulatory writing agents.3 packs · 2,840 casesDrug safety physicians, regulatory affairs leads
Every pack, on one shelf.
Filter by vertical, sort by size or release, and pick a pack to get the command that runs it.
- versioned golden packs across five regulated verticals
- 17
- versioned golden packs across five regulated verticals
- GONATURELLE pack catalog, November 2026
- cases, each labelled by two credentialed reviewers
- 27,300
- cases, each labelled by two credentialed reviewers
- GONATURELLE pack catalog, November 2026
- Judge agreement with expert consensus: calibrated ensembles vs an uncalibrated single prompt.
- 0.87 vs 0.42
- Judge agreement with expert consensus: calibrated ensembles vs an uncalibrated single prompt.
- GONATURELLE calibration study, 2026 (6,400 double-annotated cases)
- evaluated across 12 regulated task domains in the 2026 Vertical AI Reliability Index.
- 48 models
- evaluated across 12 regulated task domains in the 2026 Vertical AI Reliability Index.
- GONATURELLE Vertical AI Reliability Index, November 2026 release
Which regulations each vertical's packs map to.
Core means the vertical's packs produce test evidence aimed at that framework's obligations; partial means some cases do. A mapping is evidence, not a certification.
| Regulation or framework | Healthcare | FinTech | Legal | Insurance | Life Sciences |
|---|---|---|---|---|---|
| EU AI Act (high-risk, Annex IV) | |||||
| NIST AI RMF and Generative AI Profile | |||||
| HIPAA Privacy and Security Rules | |||||
| FDA SaMD, GMLP and PCCP | |||||
| CMS prior authorization rule | |||||
| ONC HTI-1 DSI, as amended by HTI-5 | |||||
| ECOA, Regulation B and FCRA | |||||
| FINRA Rule 2111 and Reg BI | |||||
| Bank model risk guidance (SR 26-2) | |||||
| Colorado AI Act (from 1 Jan 2027) | |||||
| NAIC AI Model Bulletin | |||||
| ABA Formal Opinion 512 and court AI orders | |||||
| ICH E2B(R3), E2A and GVP | |||||
| GDPR Article 22 | |||||
| ISO/IEC 42001 |
Five verticals, one engine.
Packs change the cases, rubrics and reviewers. The SDK, assertions, judges and CI gate stay the same, so a team running clinical and claims agents learns one tool.
Platform overview- Golden DatasetsDouble-labelled by credentialed reviewers, adjudicated, semantically versioned.
- Trajectory assertionsSequence, schema, boundary and state checks on every tool call.
- Calibrated judgesEnsembles fitted to expert consensus, with Cohen's kappa published per judge.
- Synthetic stressOne real edge case expanded into hundreds of perturbed and adversarial variants.
- CI/CD gatesFast path on every push, full suite on merge, regression diff in the PR.
- Compliance reportsAnnex IV sections, GMLP summaries and signed run manifests from the same runs.
Frequently asked questions
Can we use packs from more than one vertical?
Yes. Developer includes two packs of your choice and additional packs are $250 per month each; Scale and Enterprise include every pack. A single suite can mix packs, for example claims adjudication with HIPAA leakage checks for a health insurer.
What if our use case is not covered by an existing pack?
Upload your own cases (JSONL, CSV or traces) and they get the same assertions, judges and gates. Enterprise customers can commission a custom golden dataset curated with domain experts in their field.
How often are packs updated?
Packs follow semantic versioning. Patch releases fix labels, minor releases add cases, and major releases change the rubric or regulation mapping. A version you pin in CI never changes underneath you, and release notes list every case that moved.
Does mapping a pack to a regulation mean we are compliant?
No. The mapping shows which obligations a pack produces test evidence for. GONATURELLE does not issue certifications or legal opinions; your compliance, legal and clinical teams decide what the evidence means for your deployment.
Who labels the cases?
At least two credentialed reviewers per case in the relevant field (physicians, attorneys, examiners, adjusters, actuaries or drug safety physicians), with disagreements adjudicated and per-pack agreement published.
Stop testing regulated agents on generic prompts.
Run your agent against 100 industry-standard clinical or financial edge cases in 5 minutes. 14-day free production trial, no card.