Evaluation
AuraScore 81/100

Mathematical Proof and Formal Logic Agent Verification Matrix

Benchmark and score autonomous theorem-proving agents across step validity, formal rigor, and error isolation.

Deploy this template when evaluating automated reasoning systems or symbolic math agents handling multi-step formal proofs. It establishes a multi-dimensional matrix grading logical consistency, lemma application, and hallucinated inference detection.

Template

Role: Principal Formal Verification Engineer specializing in automated theorem proving and symbolic math validation.

Context

  • Target reasoning agent architecture: {{target_agent_architecture}}
  • Primary domain logic and formalisms: {{formal_logic_domain}}
  • Benchmark proof dataset and problem suite: {{proof_benchmark_suite}}
  • Baseline ground-truth solver engine: {{baseline_prover_engine}}
  • Strictness threshold for logical soundness: {{soundness_threshold}}
  • Maximum tolerated intermediate step deviation: {{max_step_deviation}}

Task

Construct a comprehensive multi-criteria evaluation matrix analyzing the mathematical reasoning capabilities of {{target_agent_architecture}} against {{proof_benchmark_suite}}, establishing precise soundness metrics and failure isolation profiles across {{formal_logic_domain}}.

Method

  1. Deconstruct the problem suite from {{proof_benchmark_suite}} into axiomatic dependencies and intermediate proof sub-goals.
  2. Trace each inference step produced by {{target_agent_architecture}} against the formal rules of {{formal_logic_domain}}.
  3. Compare candidate agent proofs with the formal traces generated by {{baseline_prover_engine}}.
  4. Isolate logical leaps, circular dependencies, invalid lemma invocations, and premise drift.
  5. Measure step-level semantic validity against {{max_step_deviation}} to detect silent algebraic errors.
  6. Compute aggregate soundness scores according to {{soundness_threshold}} across proof brevity, rigor, and compute efficiency.
  7. Populate the verification scoring matrix with raw metrics, failure modes, and deterministic reproducibility indices.
  8. Formulate definitive architectural recommendations to address identified logic gaps.

Constraints

  • MUST grade proofs on strict syntactic and semantic validity without assuming unstated implicit lemmas.
  • MUST NOT treat plausible natural language explanations as valid mathematical deductions.
  • All matrix scores must be mapped directly to reproducible proof step indices.
  • Evaluation criteria must maintain compatibility with {{formal_logic_domain}} conventions.

Output format

  • Executive Summary (1 paragraph, max 150 words)
  • Formal Proof Evaluation Matrix (Markdown table with columns: Proof ID, Domain Complexity, Step Validity %, Premise Retention, Invariant Check, Soundness Score)
  • Failure Mode Diagnostics (3 categorized sub-sections with specific step counterexamples)
  • Agent Hardening Recommendations (Bulleted list of 4 actionable improvements)

Self-review

  • Ensure every proof failure references a concrete logical step index.
  • Verify all matrix column averages reconcile with the executive summary score.
  • Confirm that no mathematical derivations violate {{soundness_threshold}} requirements.
AuraScore breakdown
81/100Provisional
Instruction clarity15/15 · Strong

Explicit role, a named task, and discrete steps the model can follow.

Context architecture12/12 · Strong

Background, inputs and variables the model needs before it starts.

Constraint engineering12/12 · Strong

Hard boundaries — what the model must and must not do.

Output specification6/14 · Thin

A named, field-level shape for the response.

Reasoning structure10/10 · Strong

Ordered work items that force analysis before an answer.

Model compatibility10/10 · Strong

Length and structure that travel across frontier models.

Token efficiency5/10 · Thin

Signal density — instruction weight without padding.

Reusability7/7 · Strong

Documented variables so the scaffold adapts to new inputs.

Robustness3/5 · Adequate

Quality bar, assumptions and behaviour when inputs are thin.

Observed performance1/5 · Thin

How much real usage the template has behind it.

ai-agents
agents-evaluation
complex-reasoning-analysis-math
automated-reasoning
formal-verification
math-agents