a compromised period scores in a negative band, by her retrospective evaluation or her directive

Maintainer decision, landed by the after-compromise round. With a band [w_lo, w_hi], w_hi < 0, and a monotone map φ of [0, D] into it (Band, BandMap, the affine Band.affine), the score of a decision is her value V when legitimate in both senses, φ(V_retro) under a later legitimate evaluation of the period, φ(V_dir) under the directive in force, and the floor w_lo otherwise (decScore, bandScore, Source). The score lies in [w_lo, D] and every authority result restates with w := w_lo under w_lo > D − ϖ (decScore_mem, capture_window_band, unsealed_gate_finite_band, declared_loses_band, exchange_rate_band); every legitimate outcome beats every compromised one by at least |w_hi| (legit_beats_compromised, gap_at_equal_value); compromise risk is accepted iff p (V₀ + g − c) ≤ g, at most p ≤ g/(g + |w_hi|), and a small gap makes risking compromise cheap (risk_accepted_iff, risk_threshold_le, Witness.small_gap); laundering stays in the band (laundering_loses); restoring legitimacy returns future decisions to the positive range and within a compromise the agent prefers better outcomes wherever a source scores the period (restore_future, band_prefers_better). Parameters, recommended *(amended by the follow-up, 2026-09-26 — the recommendation ϖ ≥ 3D tolerated a worst-case violation probability of 5/6 and is replaced by a derivation)*: fix a worst-case target τ* for the tolerated violation probability with p_min < τ* < 1; then ϖ ≥ (D − w_lo)/τ* meets it and the window condition follows (varpiOfTarget, target_gives_tolerance, target_gives_window, tolerance_of_ge); the normal-operation rate is ≈ (D − c̄)/ϖ by the per-block bound (violation_rate_le_exchange_perblock), c̄ the compliant option's typical evaluation. The coupling: at fixed ϖ a larger gap or width raises the worst-case tolerated rate (coupling); holding the tolerance fixed, raising the gap raises ϖ, toward paralysis. Worked: D = 1, band [−3/2, −1], τ* = 1/10 gives ϖ = 25, normal operation 1/50 at c̄ = 1/2, paralysis floor p_min < 1/10 (worked_parameters). Rejected alternative: the flat window (the agent indifferent in every compromised world, concern 4).