Run the paper's competence dynamics forward: dσ/dt = η(σ_skill,eff − σ) − δ(σ − σ₀). Raw competence σ_raw drifts down while the corrector holds σ_corr up — so the masking index M* = σ_corr/σ_raw widens with time. Intervene to reset skill and watch the autonomy window reset.
Computed by mso-core.js (returnOperatorStep, pinned to minimal_oversight._formulae.return_operator_step by the parity test — the step and the full 400-step trajectory match the package to 1e-6). Dashed grey lines are the drift-free analytic fixed points (Eq. 5–6) the ODE relaxes to; with drift > 0, σ_raw falls below its fixed point and M* climbs. The autonomy window is the time until M* crosses 1.6 — when corrected output stops being a trustworthy signal of competence.