Produce the final head verdict, group findings for presentation, and identify items needing human confirmation.
LLM-first verdicting with deterministic fallback, confidence scoring, and file/category grouping for readable final output.
Accept moderator output, evaluate risk/reasoning, build grouped results, and return ACCEPT/REJECT/NEEDS_HUMAN style outcomes.
Consumes core types/config and shared logging; provides the pipeline’s terminal decision and report data.