v3.15.13: Large-project experience search, real transfer gate, more honest results

What's New

• Experience search works on large projects. In the published package, ReasoningBank warm-up and experience consolidation failed once a project had more than 10,000 stored experiences. Both work again (#1237).
• Cross-domain transfers are checked again. The published package had replaced the coherence check with a stub that approved every transfer. The real gate is back.
• "Not measured" instead of guessed results. When there is no evidence, these now say so:
  • quality metrics, with no more passing defaults such as coverage 80;
  • accessibility audits, which no longer invent findings without a browser;
  • DAST scans, which are never clean when blocked or partial;
  • registry lookups and outdated-package checks;
  • dotCover coverage, with no more file-N placeholders.
• Code search finds code indexed with code_index, and semantic results say whether they came from vector or text matching.
• task_status can be polled until a task finishes, without tripping loop detection.
• aqe fleet run --parallel splits the work across workers, and --format json is new.

Getting Started

npx agentic-qe init --auto

See CHANGELOG for full details.