How Compass works
The product uses rules-first evaluation with optional AI assistance for text quality and classification hints.
Rules-first pipeline
Every evaluation follows the same macro logic: intake, classification, follow-up questions, scoring, concern aggregation, result framing, and optional refinement.
Deterministic rules own labels, concern levels, and recommendation severity.
Where AI fits
AI may help summarize the free text, infer likely tags, generate cleaner headlines, and tailor investigate questions.
If AI fails, Compass still produces a complete result through deterministic templates and scoring.