axiom
Neuro-Symbolic SymPy Conservation-Law & Formal Verification Compiler for Hermes Agent.
LLMs reason probabilistically via softmax over tokens and routinely violate hard conservation laws (double-entry accounting, Kirchhoff current sums, probability simplex normalization $\sum p_i = 1$, thermodynamic bounds, and SI dimensional homogeneity).
axiom wraps Hermes Agent's llm_execution and tool_execution middleware in a fast, zero-eval() AST-to-SymPy + SciPy Linear/MILP constraint solver:
- Exact Counterexample Witness Generation: When an equation system or resource allocation is inconsistent (
UNSAT) or violates an invariant,axiomcomputes the exact minimal-perturbation feasible assignment $x^*$ and feeds the constructive counterexample back to the model. - SI Dimensional Matrix Homogeneity ($[M^a L^b T^c I^d \Theta^e N^f J^g]$): Proves dimensional consistency via exact matrix rank / nullspace algebra.
- Conservation-Law Audit: Audits financial ledger flows, mass/energy balance, and probability distributions in $< 5\text{ ms}$.
- "Less is More" UX: Appends a whisper-quiet
Q.E.D.badge (::axiom) expandable to inspect the symbolic proof certificate.
README from the reviewed commit 2e53a66; it updates when the author re-pins.

arjunbagga-xyz