Eval Suite & Diagnostics

# Statistical Benchmark Suite (`tacet-eval`)

`tacet-eval` measures tool selection accuracy across English and Turkish prompt suites under reproducible benchmark conditions.

• Sign Test Metric: Calculates exact p-values (e.g. p = 1.0) over paired test suites to verify statistically significant model improvements.
• Noise-Free Evaluation: Disables sampling temperature and random noise during benchmark runs for 100% reproducible results.
• Detailed Fingerprinting: Benchmark reports record model path, tensor quantization, hardware device, wall time, and active tool catalog fingerprints.

# Diagnostic Command: `tacet why`

The `tacet why` command analyzes trigger matching, term boundary scoring, and tool budget allocation in milliseconds without initializing model weights:

tacet why "Dolar kuru şu an ne durumda?"