auto-0.4b-2 iteration 1: shorter card and iteration-1 naming (weights unchanged)

auto-0.4b-2 v3: distillation from Auto 3B, cross-paired KD data, weight averaging

Show the exact user request alongside every measured tool probe

Add 24 measured skill and MCP probes; display ModernBERT-large as the base model

Align paired accuracy statistics with the documented deny-on-tie threshold

Publish evaluated auto-0.4b-2 checkpoint and reproducibility artifacts

initial commit