v0.1.5

Added

• [tool.longtracer] config support in pyproject.toml — set project-level defaults (project name, backend, threshold, verbose, log_level) without code changes.
• Config priority chain: Code args > Environment variables > pyproject.toml > Built-in defaults.
• verify_batch() method on CitationVerifier — verify multiple response+sources pairs in one call using parallel ThreadPool execution.
• verify_batch_async() — async wrapper for verify_batch().
• check_batch() top-level convenience function — one-liner batch hallucination check.
• Improved HTML trace report:
  • Per-claim source diff — side-by-side view of LLM claim vs best source evidence, color-coded by status.
  • Summary stats bar — visual pass/fail/hallucination breakdown with colored segments.
  • Inline score bars on each claim row.
  • Click-to-expand claim detail with STS, entailment, and contradiction scores.

Changed

• LongTracer.init() now reads defaults from [tool.longtracer] in the nearest pyproject.toml.
• CitationVerifier reads threshold from pyproject.toml config when not explicitly passed.
• parallel_pipeline.py now passes best_source, contradiction_score, and score fields to the eval_claims trace span for HTML diff rendering.