mghafiri/qwen3.5-0.8B-decision-model

Add head-to-head comparison with TypeSafe Jev on unseen scenarios

Train 10 epochs; publish best epoch with eval report

initial commit