[<< BACK TO PROJECTS]
Parametric Insurance + AI Evaluation2025 · Research Prototype

# ParaEval

A workbench to adjudicate parametric trigger decisions against multi-source evidence. It makes the decision path explicit and expandable so reviewers can trace Situation -> Task -> Action -> Result for each case.

Parametric InsuranceZodSQLiteLLM ExtractionEvaluationNext.js

Overview

ParaEval is an evaluation workbench for parametric insurance trigger decisions. It normalizes heterogeneous evidence sources, applies deterministic decision algorithms, and surfaces disagreement as first-class basis risk. Heterogeneous data sources (gauges, satellite imagery, weather APIs) are normalized into a common schema for comparison and adjudication. A deterministic decision algorithm evaluates trigger conditions against normalized evidence, with Zod schemas mirroring future Pydantic models for a clean TS-to-Python bridge. Near-threshold disagreements across sources are surfaced as first-class basis risk, highlighting why sources diverge and how that affects confidence. The engine is regression-tested against golden cases with 31 unit tests covering algorithm branches.

Problem

Policies trigger on index values (gauges, satellite, APIs) rather than on verified loss. Near-threshold disagreements across sources create ambiguity and hidden basis risk without a shared reasoning frame.

Approach

Normalize heterogeneous sources, apply a deterministic decision algorithm, surface disagreement as first-class basis risk, and regression-test the engine against golden cases. The contract layer (Zod) mirrors future Pydantic models for a clean TS<->Python bridge.

Impact

  • * Transparent rule-trace and narrative output suitable for review memos and audits.
  • * Explicit basis-risk classification highlights why sources diverge and how that affects confidence.
  • * 31 unit tests cover algorithm branches to prevent silent regressions as cases and rules evolve.

Architecture

ParaEval is an evaluation workbench for parametric insurance trigger decisions. It normalizes heterogeneous evidence sources, applies deterministic decision algorithms, and surfaces disagreement as first-class basis risk.

Layer
Implementation
Purpose
Data ingestion
Multi-source normalization pipeline
Heterogeneous data sources (gauges, satellite imagery, weather APIs) are normalized into a common schema for comparison and adjudication.
Decision engine
Deterministic trigger algorithm + Zod contracts
A deterministic decision algorithm evaluates trigger conditions against normalized evidence. Zod schemas mirror future Pydantic models for a clean TS<->Python bridge.
Basis risk analysis
Disagreement detection + classification
Near-threshold disagreements across sources are surfaced as first-class basis risk, highlighting why sources diverge and how that affects confidence.
Regression testing
Golden case suite + 31 unit tests
The engine is regression-tested against golden cases with 31 unit tests covering algorithm branches to prevent silent regressions as cases and rules evolve.
Output
Rule-trace + narrative memo generation
Transparent rule-trace and narrative output suitable for review memos and audits. Each decision path is explicit and expandable for reviewers.