Release v0.3.0 · ENDEVSOLS/LongTracer

Added make benchmark — opt-in benchmark harness (benchmarks/evaluator/run.py) that runs verify_case with real weights and reports per-label precision and recall, confusion counts, false-alert rate and evaluator errors, every rate with its denominator and a Wilson 95% interval. Not run in CI. benchmarks/evaluator/dataset.json — 140-case evaluator benchmark dataset (14 categories, calibration/held-out split, leakage-checked, pinned models and thresholds), generated by benchmarks/evaluator/dataset.py. Labels are