mandrel-bench: v0.3.0

0.3.0 (2026-06-18)

Added

• bench: batch-ready run orchestrator — resumable, cost-bounded loop (refs #22) (#24) (9d4d871)
• bench: drive the mandrel arm via /plan --idea --yes (headless, fresh Epic per run) (#28) (81d5093)
• bench: make mandrel-arm runs clean and repeatable (#27) (4aaf208)
• restructure results/ into per-cohort directories and add a generated zero-dep results.html dashboard (#17) (#19) (dfe8c13)
• results: first N=8 baseline cohort — mandrel@1.72.0 / claude-opus-4-8 (refs #23) (#29) (5100d9d)

Fixed

• bench: render the value-add report over the full cohort store (resume-safe) (#31) (e564b3d)
• bench: sanitize GITHUB_TOKEN before gh in resetSandboxBaseline (#30) (a50cfe5)