v1.5.1
HaltEvidence structured recovery on halt, LLM-as-judge hybrid escalation, cross-encoder reranker, bidirectional AggreFact benchmark, streaming window sweep ablation, CI regression suite, domain cookbooks, marketing language cleanup, license FAQ, badge updates. 810 tests passing. Regression suite runs in <1s without GPU. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>