The Assumed Fast Variability and the Warning Rate of a Frozen Predictive-Regime Monitor on a Verified Fold Transition: A Pre-Specified Paired Sweep of the Fast Term's Amplitude

In this prospectively specified simulation study, the assumed fast observation term of a verified fold generator was varied over seven amplitudes, from zero to the published 2.80 Sv, with the slow-state trajectories, the frozen predictive-regime monitor and everything else held fixed, so that every difference between amplitudes is a paired difference on the same noise path. In the primary cell (the companion validation study's headline cell with fresh seeds: an assumed slow-state amplitude of 1.0 Sv, a 600-month forcing ramp, and the baseline a 67-year reanalysis record affords), removing the fast term raised the frozen monitor's warning rate from 3 of 195 eligible runs (1.5%) to 10 of 195 (5.1%), a paired increase of +0.036 (Newcombe 95% interval +0.004 to +0.075; discordant counts 8 and 1): the study's one confirmatory comparison, read under its registered rule as supporting an increase, with a lower limit close to zero. Warning remained rare. The monitor is in the registered band little or no warning at every amplitude in both cells; its false alarms on the unforced baseline rose with its warnings (+8.7 percentage points, +3.3 to +14.5); and against a time-matched null, each ramp run's unforced twin scored over the ramp run's own warning window, it discriminated at no amplitude in either cell. Of three secondary checks, P1 (the zero-amplitude warning rate's Wilson upper limit below 20%) holds at 0.051 (0.028 to 0.092); P2, that removing the fast term would bring the mechanistically matched lag-1 configuration under the study's false-alarm ceiling, was not borne out, its false alarms rising from 25 of 184 to 85 of 184 (ceiling 9); and P3, that the fresh seeds at 2.80 Sv are compatible with the published held-out run, passes (0.015; 0.005 to 0.044, overlapping the published 0.008 to 0.051). The generator, its verification, the design constants, the frozen monitor pair (a lag-2 diagonal representation requiring four consecutive positive refits) and the two classical reference detectors are the validation study's, unchanged; only the fast term's amplitude varies, over 0, 0.25, 0.5, 1, 1.4, 2 and 2.8 Sv. Because each attempt's seed feeds the state and the fast term from separate draws, every amplitude sees the same state trajectory, burn-in decision, event time and eligibility, and the fast term at each amplitude is the published one scaled; a pairing test on a reserved seed range verified this before registration. Two cells of 200 accepted attempts were run with fresh seeds: the primary cell above, and the 0.5 Sv cell on the same layout as a sensitivity comparison, whose paired change from 2.80 Sv to zero (+0.010; -0.015 to +0.039) does not resolve the direction of the change. Three populations carry the rates, each with its denominator printed: the eligible ramp runs, the event-free null runs, and their intersection, on which the warning rate and the matched null rate are two readings of the same detector over the same calendar months on the same noise path, differing only in the forcing; their paired difference carries Newcombe's score interval for paired proportions and the exact McNemar p-value. The question, the grid, the cells, the seed ranges, the populations, the one confirmatory comparison, the secondary checks and the sentence each outcome licenses were committed to a version-controlled record before any attempt of the study's seed range was generated, and the paper states that design as carried out. Every number in the paper is emitted from the committed results by a tabulation step, and an independent checker written without reference to the scoring code recomputed every quantity on its own denominator and regenerated attempts from the generator at every amplitude. The study tests, in the one form its design can test it, the reading the validation study's diagnostics suggested but could not test: that the assumed fast term hides the fold's signature from the detector. The effect found is small. With the fast term removed entirely, the monitor warns before about one eligible transition in twenty, where with the published fast term it warned before about one in sixty-five. The study establishes the effect of the fast term's amplitude on this frozen detector within this generator; it does not identify the mechanism of that effect, does not establish useful warning, does not re-select a detector, and says nothing about the fast variability of real indicators. The two reference detectors stayed in the same band at every amplitude, with descriptive discriminations from their matched nulls at the lowest amplitudes in the sensitivity cell only; no detector is promoted on their account.