Evaluation Report (1,000 Milestone) Model: jev-latest (v1.13.0) Sample: 1,000 Decisions

JEV Deterministic Decision Benchmark (JEV-DD-1.0)

Comprehensive diagnostic profile across 25 reasoning families, confidence calibration, and error mechanics.

Cumulative Accuracy
81.7%
817 / 1,000 decisions
Static Relational Logic
98.4%
Flawless across 7 core families
High Confidence (≥0.95)
99.8%
492 / 493 correct (49.3% vol)
Sequential Procedures
13.2%
Core: only 2 / 26 (7.7%)
Discrimination (ROC AUC)
~0.93
Exceptional calibration

Reasoning Family Capability Spectrum

Comparing accuracy across all 25 benchmark families (1,000 decisions total: Solid = All items, Marker = Core items)

Very Strong (90-100%) Moderate (75-90%) Weak (50-75%) Extremely Weak (<35%)