# Elite Reasoning MCP

> Elite Reasoning MCP is an evidence-gated verification, selection, and measurement layer for coding agents. Every capability claim regenerates from committed raw data; retired claims are retracted, not deleted.

## Core Verification & Preparation Gateway

- [elite_prepare](https://github.com/Snehgabani/elite-reasoning-mcp#tool-elite_prepare): Start a non-trivial task. Retain run_id and obey the returned continuation after each later Elite call. Not the answer.
- [elite_verify](https://github.com/Snehgabani/elite-reasoning-mcp#tool-elite_verify): Run one check, persist evidence by run_id, then follow the returned continuation before answering.
- [elite_memory](https://github.com/Snehgabani/elite-reasoning-mcp#tool-elite_memory): Search, record, or explicitly approve scoped memory items.

## Measured Record (regenerates from committed raw data)

- [Boundary guard verifier](https://github.com/Snehgabani/elite-reasoning-mcp/blob/main/docs/E1_VERIFIER_FIX_RESULTS.md): 0% false alarms at full recall on a 113-solution labeled corpus (previously 20.4%).
- [Contract requirement extraction](https://github.com/Snehgabani/elite-reasoning-mcp/blob/main/docs/E3_CONTRACT_V2_AND_CEILING_FINDING.md): 100% verbatim requirement recall on the labeled spec corpus (previously 46%).
- [Draft selection](https://github.com/Snehgabani/elite-reasoning-mcp/blob/main/docs/E3_SELECT_CONFIRMATORY_RESULTS.md): CodeT-style best-of-8 dual-execution-agreement selection via elite_verify(check="select"); preregistered confirmatory replication on Llama-3.2-1B (140 fresh pairs) beat single-draft on all-hidden-tests-pass 15 vs 7 (exact McNemar p=0.0215) at 14.9x token cost. One model, one suite; not independently replicated.
- [Evidence gating & measurement](https://github.com/Snehgabani/elite-reasoning-mcp/blob/main/docs/PRODUCT_REVIEW_AND_MEASUREMENT.md): digest-bound evidence gating, a reproducible A/B harness with committed graded runs, and retraction + anti-fabrication CI gates.

## Full Specification

- [llms-full.txt](https://raw.githubusercontent.com/Snehgabani/elite-reasoning-mcp/main/llms-full.txt): Complete uncompressed API reference and prompt integration schemas.