test_graph_routing.py
Protects conditional graph paths such as cache-hit routing, clarification routing, retrieval/news selection, calculator routing, and guardrails outcomes.
Tests and Evals Map
A static guide to what the repository tests and evaluation suites are trying to protect. Use the filters to focus on a behavior class.
routing.jsonl
Checks whether the router makes the right high-level decision for user intent: report retrieval, news, clarification, or direct answer.
test_merge.py
Verifies that duplicate chunks from parallel retrieval branches are removed by a stable document key.
test_retrieve_decision.py
Checks the decision node behavior around retrieval, direct answers, ambiguity, and out-of-scope input.
retrieval.jsonl
Evaluates whether the RAG path can find relevant annual-report evidence for financial questions.
test_calculator.py
Protects deterministic financial math tools: generic calculate, growth rate, ratio, and average.
calculation.jsonl
Checks whether questions that require arithmetic are routed and answered with calculated values.
test_guardrails.py
Ensures empty or too-short context fails before answer generation.
test_fallback.py
Verifies that failed guardrails produce a helpful fallback instead of a broken answer.
test_citations.py
Protects citation filtering, renumbering, source previews, and superscript conversion.
boundary.jsonl
Checks boundary behavior such as unrelated questions and graceful refusal.