Prototype 01 · corgi-insurance · synthetic decision set · noindex

Which AI decisions must be a rule a regulator can run, and which legitimately stay a model?

An AIS-Program examiner does not ask for a SHAP plot. They ask you to run the rule that denied a claim or declined a risk. This tool takes a list of AI-influenced underwriting and claims decisions, classifies each into the route an examiner expects, and flags the ones that are compliance-bearing yet currently model-made. The decision set below is synthetic, seeded from Corgi's public line list; your real inventory is phase 1 of the engagement.

SYNTHETIC DATA: a Corgi-shaped decision set built from the public commercial-specialty line list. No Corgi-internal decision, model, or number ships here.

Try a scenario: one click loads the inventory and re-runs the classifier

Fine-tune (optional): scope the inventory

Line of business

Decision function

Decision	Line	Function	Today	Examiner route	Why

Sources & method

Synthetic decision set: a Corgi-shaped inventory seeded from the public commercial-specialty line list (D&O, E&O, Cyber, Tech-and-AI liability, EPLI, Workers' Comp) on corgi.insure. The decisions, the route mappings, and every count are ILLUSTRATIVE; no Corgi-internal decision or model is represented. Corgi's real inventory is gathered in phase 1.
The route taxonomy (must-be-a-rule / legitimately-a-model / human-at-the-edge) is the classifier's argument, grounded in the AIS-Program exam request list: which AI systems are in production, what decisions they influence, evidence of ongoing testing, monitoring over the reporting period (NAIC Model Bulletin).
Why claims-triage flags: Colorado SB21-169 / Reg 10-1-1 bars ECDIS-driven unfair discrimination across rating, underwriting, and claims, and asks for the protected-class testing, not the model's internals.
The honest bound: this is not anti-model. The clean-model rows say so. The claim is architectural: which specific decision should be a readable rule in front of the model, not which models to delete.