chore: release 3.11.0

[joshuawheelock]

[3.11.0] - 2026-06-24

Added

• TUI global search: Press / in the TUI to search the projected global memory index interactively, with grouped result metadata and a focused generic detail preview for the current hit.
• Evals replication statistics (Outcome 5, 1/2): Added a pure aggregateReplications aggregator that combines K replicated A/B comparisons of the same scenario into a ReplicationReport, reporting per-dimension mean lift, sample standard deviation, and a t-statistic against the K=5 one-tailed alpha=0.05 threshold so lift can be reported with statistical confidence instead of single-point estimates. Token-efficiency replications scored N/A are excluded and counted separately.

Changed

• Evals token-efficiency equivalence gating (Outcome 4): Token efficiency is now reported only when both arms are output-equivalent (both produced every expected file and both reached a structural-assertion score >= 0.8); otherwise it is reported as N/A so the ratio never compares the cost of producing different things. The Jumbo arm's lifecycle adherence rate is now surfaced separately in the protocol-adherence dimension rather than averaged into quality lift, keeping quality scores symmetric across arms.
• CI: Upgraded GitHub Actions workflows to Node.js 22 (actions/checkout v4 → v7, actions/setup-node v4 → v6), clearing the Node.js 20 runner deprecation warning. Contributed by Allan Kimmer Jensen.