More control over your dispatch

Fixed

• SimBrief-supplied routes no longer fail a blank-route readback. Explicit routes still must match. The approved runway is selected after the departure-time prompt resets it.
• Generation failures return clicked and retryable. Safe failures before Generate may be corrected and retried once; an atomic database claim prevents a second Generate click across threads or restarts. Repeating dispatch cannot erase generation history.
• Overlapping SimBrief dialogs are handled from the front. Unknown warnings stop for review.
• Connection durations are computed in a tool before recommendations. The prompt avoids redundant confirmation of settled flights and labels airport field elevation explicitly.
• Briefings expose SimBrief fuel/range/weight failures and preserve provider messages. Infeasible plans withhold thrust-setting guidance. A generated OFP is distinct from a feasible flight; the observed KLAX–NZAA plan had negative landing fuel and a range warning.

Added

• Guided full-throttle N1 calibration and percentage targets. Saved readings require a simulator version, exact aircraft/engine/conditions/runway match, and an age of at most seven days. No engine model or supported-aircraft claim changes.
• Dated arrival-target calculations using sourced gate-to-gate block time, with explicit timezone/date-line handling and comparison to the generated plan's arrival.
• Explicit route, runway, load and SimBrief airframe selection. The retained OFP checks the selected airframe id. Simulator configuration/capacity and its source are kept separate from the provider profile; supplied capacity uses the dispatch's explicit KGS/LBS units.
• Saved-trip listing, per-leg status/briefings, revision comparisons, approved revisions, briefing refresh with changes, and persisted conversation/approval inspection after restart.
• Per-turn FlightAware list-price measurements and Browserbase elapsed time (no invented Browserbase/model invoice costs); versioned traces and objective workflow feedback.
• scripts/review_traces.py for bounded conversation review and optional LangSmith queueing; evals/run_workflow_replay.py for real-model generation/recovery/briefing regression with controlled providers and zero FlightAware/Browserbase calls.

Live acceptance is recorded separately after deployment; see the implementation and validation report. No new aircraft-support claim is made.