ic_gen and spectral_random_velocity cite the measurement that already covered them, and the two wall-bounded providers were measured against their contract (wall-spectral-ic-2026-09-22): on stretched channel and duct grids the staged field carries the requested bulk velocity and perturbation RMS exactly, its mean equals the flux-normalized product parabola to 3e-15, the walls carry zero velocity, and the flux field the runtime reconstructs is divergence-free at round-off. They remain startup constructions: nothing here establishes sustained turbulence.models.physics.turbulence.rans is refused at validation and -rans at setup; both name the status rather than ignoring the setting. Nothing behind the selector was ever implemented - the K_Omega fields were never allocated, the transport update in FlowSolver was commented out, and the function it called was defined nowhere - so a case that enabled it aborted on a null vector at the end of the first step. It was recorded known-defective on 2026-09-18; since no implementation had ever existed, removing the dead hooks returns it to planned, with the charter for a future closure on page 57 and the removed hooks listed in src/guide.md.K_Omega fields and their catalog entries, the FIELD_AVAILABILITY_RANS flag, the two-component DM they alone used (never created), the simCtx->rans branches in the RHS, run loop, checkpoint I/O and Newton-Krylov scope check, and the k_omega capability value.-checkpoint_rans is no longer written or required. The format version is unchanged, so existing checkpoints still load; a checkpoint written now is not readable by a pre-removal binary, which requires that key.cabot and werner under it) went with the selector.known-defective to planned, with a recorded reason saying what was found and what was removed: a feature that never existed is not a feature that broke.