Speed grids (canonical) and the retired adaptive-map research lane

Canonical speed grids

Every SFINCS v3 speed-grid option is canonical in dkx.phase_space (make_speed_grid, xdot_diff_matrices, make_grids): xGridScheme 1-8 (the Landreman–Ernst polynomial grids with and without a node at x=0, the uniform grids, and the Chebyshev variants) together with xDotDerivativeScheme -2..11 for the upwinded xDot derivative pairs. The polynomial x-grid kernel consumed by the collision operators lives in dkx.xgrid. Parity is pinned by Fortran goldens in tests/test_output_h5_xgrid_schemes_parity.py and the phase-space unit tests.

Retired research lane

The differentiable adaptive-map primitives (monotone maps x = g_theta(eta) with positive Jacobians, the opt-in xGridScheme = 50 construction, and the mapped-grid transport-evidence workflow) were research material hosted by the legacy discretization/workflows.mapped_xgrid owners. They were deleted with the legacy pipeline; the lane is preserved in git history and may return through Open research lanes gates as a differentiable-grid optimization workflow on the canonical stack.