What problem does it solve? Performance and no-behavior-change claims about the FutureFin stack are often made by eyeballing instead of measuring. This Skill provides verified curl recipes, log-grep patterns, SQL queries, and three shipped shell scripts to time endpoints, prove projection-cache HIT vs MISS, check gzip payload sizes and density point counts, diff projection responses before and after a change, and read row counts and migration state straight from Postgres. ## Core Features & Use Cases - Endpoint timing and cache verification: Timed curl GETs against /v1/projection/series, RUST_LOG axes for cache telemetry (HIT/MISS/compute ms), and the smoke-projection-cache.sh flow that shows MISS→HIT after a real mutation. - Payload and density measurement: Raw-vs-gzip size checks, monthly vs hybrid density point counts, history/cash-flow window caps, and the MCP catalog context-cost recipe measured from the frozen fixture. - DB-level diagnosis: SELECT-only queries through the single 3.0.0 container's Unix socket (psql -h /var/run/postgresql), covering _sqlx_migrations state, sessions, pending users, expired liabilities, and EXPLAIN ANALYZE patterns. - Use Case: After refactoring the projection engine, run projection-diff.sh --save before the change and --compare after it to produce byte-level evidence that KPIs, milestones, and shared month_index values did not diverge. ## Quick Start Ask the AI to measure whether the FutureFin projection cache is working by timing /v1/projection/series twice per density and grepping the container logs for cache HIT and MISS lines.
