FAQ · P1

Token-efficient REA (right-sized Evidence)

Measure tokens per verified finding — not tokens per vibes. Efficiency never means inventing summaries.

Checklist

Prefer summary / selected views in 6.4+ for Evidence, JS graphs, and dossiers before dumping raw trees into the model.
Use REA_MCP_INPUT_SCHEMA_PROFILE=compact and/or domain-split so tools/list fits strict clients (schema size FAQ).
Keep raw dumps on disk; pull slices into context only when needed.
Reuse sessions / snapshots — avoid re-analyzing the same binary cold every turn.
Ask bounded questions (“list imports of module X”) instead of “explain the whole app.”
Retain Evidence files; quote them. Do not strip limitations/unknowns to look confident.