A post-decision intervention gate in confidently committed high-ambiguity decisions.

Pre-registered reanalysis testing a post-decision intervention gate: whether confident decisions on objectively ambiguous, correct trials are revisited by a post-decision layer in a way that standard accounts of changes of mind and post-decision control do not predict. One gate, two arms. Arm 2, the powered confirmatory arm, is a next-trial caution-residual test on a pool of four Confidence Database studies. Arm 1, un-parked in this version, is a within-trial reversal test on van den Berg et al. (2016), using data obtained from the corresponding author. Outcome across the family. Arm 2 (H4) is not supported and the pooled estimate is reversed, concentrated in the no-feedback datasets, with the confidence-guided-control baseline not recovered. Arm 1a and Arm 1b on van den Berg are structurally inconclusive: the authors' analysed sample contains four participants against a pre-registered participant floor of eight, so the participant-level test cannot reach 0.05. Arm 1b's one difficulty-controlled contrast leans the standard accumulator, and the verdict holds across the 400 ms primary window and the 326 ms, 500 ms, and per-subject t_pip robustness windows. With no primary positive, the pre-registered Holm decision over H1a, H1b, and H4 changes no verdict. The result does not refute the intervention-gate account; it leaves the standard accounts as the parsimonious default and the gate unresolved at decisive resolution. This version (v0.8) corrects the Arm 1 dependent variable from coc (confidence change) to cod (choice reversal), following a standalone arbiter that found the dataset readme labels swapped relative to the data, confirmed by the data author. It also gates the Arm 1a verdict on the pre-registered participant floor. See ERRATUM.md. No data is redistributed.