Skip to content

Ticket 030: Fidelity v2 Core - Layered Wind and Sub-Segment Sampling

Status

Implemented in May 2026.

Goal

Improve model fidelity without breaking v1 estimator contracts.

Current Gap

This ticket is complete. Fidelity v2, layered wind, and sub-segment sampling are available through mission YAML, scenario YAML, CLI flags, and Python APIs. They are separate controls: fidelity v2 enables neither layered wind nor straight-leg sub-segment sampling by itself.

Scope

  • Add LayeredWindProvider.
  • Add optional fixed sub-segment sampling for long legs.
  • Add deterministic sampling policy and compatibility mode for v1 behavior.
  • Add independent schema controls for fidelity and maximum segment length.

Acceptance Criteria

  • Users can choose fidelity and layered/sub-segment behavior independently.
  • Existing v1 golden outputs remain stable unless explicitly running the v2 mode.

Superseded. The second criterion was met by leaving unset max_segment_length_m on a zeroth-order rule that sampled each leg once, at its departure end. That kept fixtures stable but understated energy whenever the wind built along a leg — measured at 30 % on a 13.7 km leg through a routine gradient. max_segment_length_m now resolves to a 500 m default and every leg is sampled at sub-segment midpoints. Fixture stability is no longer a constraint on the default integration.

Integrated Surfaces

  • Mission YAML configures estimation.fidelity, estimation.wind_layers, estimation.max_segment_length_m, and estimation.min_groundspeed_mps.
  • Scenario YAML configures the same estimator controls through initial_conditions.
  • The estimate CLI exposes --fidelity, repeated --wind-layer, and --max-segment-length-m options.
  • scenario execution honors scenario initial conditions and can use mission asset wind providers when no explicit scenario wind is set.
  • Layered wind and sub-segment sampling compose with energy, geofence, landing-zone, terrain, wind-grid, and result-envelope behavior.

Out of Scope

  • Weather grids.
  • Stochastic sampling.
  • Turn dynamics.