GymGov · Certification Governance Architecture

Separating training from certification to verify reasoning — in the age of AI.

This proof of concept runs a real Epistemic Learning Object (clinical reasoning — chest pain) through a neuro-symbolic pipeline: an LLM proposes, a deterministic ontology engine adjudicates. Train in the Gym, get certified in the Arena, govern the truth layer in the Studio.

🏋 Learning Gym

Socratic training

Formative dialogue that reveals hints and coaches your reasoning. No stakes.

⚔ Certification Arena

Adversarial certification

The same engine, strict. Defend your reasoning; the engine owns the verdict.

✦ Studio

Truth-layer review

Faculty browse, approve, and annotate the ontology-grounded knowledge base (Axis-B).

📋 Label Static

Blind faculty review

Verdict a frozen answer blind, then see the engine's grade + named-rule rationale and rate its soundness.

How it works

  1. An LLM extracts and proposes the next move — it never sets the grade.
  2. A deterministic CRO engine adjudicates against the truth layer (SNOMED/UMLS-grounded).
  3. Every asset carries provenance + an epistemic weight, capped until faculty-approved.