7. Testing the whole quote

The goal

Verify the state and policy contract without confusing a skipped provider evaluation for test evidence.

Deterministic coverage first

test/quote-graph/quote-graph.spec.ts covers the rating engine’s determinism, tier ordering, and lender rule; a scripted full quote through driver capture, trim disambiguation, vehicle use, history, coverage, adjustment, and acceptance; a rejected suspended licence; and the idle-window restore policy. Model replies come from a scripted gateway, so no provider is needed to establish those contracts.

Then run the semantic scenario

quote-graph.scenario.json defines a fifteen-turn conversation: driver collection, trim disambiguation, vehicle use, insurance history, a lender rule that blocks liability-only coverage, tier presentation, a deductible what-if, and final acceptance. The E2E evaluator logs each input, response preview, semantic-judge result, and final persisted state check.

npm run test:quote-graph   # deterministic; the E2E file is skipped
npm run test2:quote-graph  # sets USE_ENV=1 KEEP_SESSION=1

The E2E test is skipped unless USE_ENV=1 is set; with it set, it needs working provider credentials and uses the configured session store. It is evidence only when it actually runs and completes; a skipped test is not a successful provider evaluation.

Next

Return to the QuoteGraph tutorial overview or explore DecisionHotelGraph.