Order gated by critical slowing does not beat free-energy signals: a four-dataset negative head-to-head, and why the construct was not testable on existing survey data

VERSION 2 (2026-07-18) — This version corrects every Study 2 number in Version 1 (2026-07-01). The conclusion is unchanged; the numbers were not. A correction note included in this version gives the full account, including how the error was found. — Abstract — Constructs describing the order or coherence of experience are repeatedly proposed as explanations of well-being and of psychological collapse. We test whether one such operator earns its keep, and report two negatives. Study 1 (survey data). In every reachable existing dataset, an order construct built from comprehensibility / manageability / meaningfulness items is item-identical to Antonovsky's Sense of Coherence (SOC); construct distinctiveness is therefore not answerable on existing data and requires purpose-built disjoint instruments. Where a validated SOC scale exists (n = 226), a nonlinear functional form did not beat and slightly lost to the plain SOC sum out of sample (ΔR² out-of-sample = −0.037 [−0.064, −0.003] for well-being; −0.072 [−0.120, −0.027] for self-rated health). Study 2 (intensive longitudinal data). On four open multi-case ESM/EMA datasets (total N = 273 persons), a cross-channel configural order operator gated by critical slowing was tested against a prediction-error / free-energy comparator handed all shared signals first (level, variance, lag-1 autocorrelation, order, prediction error). In the powered dataset (N = 40 clinical cases; Fisher et al. 2017), the gate's increment beyond the comparator was +0.0005 pooled R² (person-fixed-effects, 3,262 windows) — negligible. The predicted direction ("rigid order is pathological") was not supported: the gate coefficient was positive in only 33% of persons in the powered set and in 52% / 42% in the two corroborating sets — inconsistent across datasets. The shared early-warning signals did essentially all of the predictive work. Implication. For what is currently measurable, free-energy / prediction-error signals absorb the cross-channel order operator. The one structure not tested here — whether collapse is a generic intrinsic tipping with hysteresis rather than a smooth gradient — requires within-person decline and recovery arms, is preregistered and frozen (10.5281/zenodo.21366132), and remains un-run. — Changes in Version 2 — - The Study 2 analysis code was lost (absent from disk, never version-controlled, unrecoverable). The Version 1 numbers were not reproducible by anyone, including the author, and were re-specified from the written specification. This was found only because the loss forced a rebuild; no check would have caught it.- Every Study 2 figure changes; the verdict does not. Largest change: Fisher median absolute R² 0.636 → 0.034 (a quantity the paper does not interpret). Pooled gate increment +0.0027 → +0.0005; total N 297 → 273.- The phrase "positive in 100% of persons" is withdrawn; the in-sample guarantee is non-negative, which is mechanical, not evidence.- The re-specification is frozen but is NOT a preregistered analysis. The analysis code and manuscript were produced with generative AI under the author's direction; the author is responsible for the content and cannot independently derive the statistical machinery. This is a negative report. It validates nothing and upgrades no claim.