A back transition can overshoot a requested hover anchor and exhaust its finite return window. Add explicit anchored capture to our owned VTOL runtime: qualify capture-time ground/air motion, fork the owned model and controller into a bounded forecast, select a finite approach/braking schedule, and continuously revoke forecasts when source, wind, uncertainty or battery evidence leaves the admitted declaration.
The opt-in fixture IMU predictor and critically damped landing flare let the declared 330 m calm/headwind/tailwind examples complete without changing the existing capture, landing, reserve, fence or independent output-lease limits. The default estimator and legacy vehicle profiles retain their contracts. Unavailable navigation produces paired null vectors in a vehicle version-3 recording, with evidence-qualified common dispatch and native replay. Capture motion is opt-in, preserving the existing pose-only schemas used by inspection, mapping, vision and cinema.
Six planned wind names expose the same cases through native Buddy, the copied standalone CLI and the common stx workspace. Their 30 onset combinations distinguish completed routes, capture planning limits and two stops before requested wind onset; unexposed wind remains unmatched and exits 1. Reports keep actual capture epochs, model assumptions, forecast samples and invalidation evidence.
Validation: native lint, app/runtime/stx types, the docs build and roadmap/parity checks passed. The 30 planned wind/onset cases and 90 recording/readiness/recovery checks passed. Native/copied CLI results matched byte for byte, including persisted unavailable-navigation replay; legacy vehicle cases were checked against retained baseline executables. Five quad reports and the complete 14-file inspection dataset match the freshly built published-main baseline byte for byte. Browser checks cover completed and refused cases, replay timestamps, case clearing, docs links and the 320-pixel layout. Both independent reviews are clean. The complete exact-head CI workflow remains required before main publication.
Retained failures: two earlier broad local runs hit default test wall-time limits; focused reruns with an explicit test wall budget passed without changing simulated deadlines. A later unintended Buddy full-suite invocation exposed a strict inspection-export format regression. The export now keeps pose-only fields by default, with planner motion explicitly requested. The original failed and interrupted logs remain retained.
These are bounded fictional model scenarios, not an interval reachability proof, measured airframe/energy qualification, firmware SITL, HIL or physical flight acceptance. General vehicle and hardware qualification stays in #12.
Refs #87.