Intercorrelation among the psychological variables violated the independence assumptions of the original correlation-matrix analyses, and dependence-aware reanalyses substantially diminish the founding excess-correlation results.

Reported as a concession against interest from inside the program, which raises credibility; not yet published as a standalone, independently reviewed reanalysis of the founding datasets, which caps it.

Simulations from within the research program itself have demonstrated that high intercorrelations among psychological variables artificially inflate the significance counts used in the original matrix studies. This dependence critique directly undermines the statistical foundation of the early positive reports. The strongest opposing consideration is that this reanalysis currently exists primarily in a doctoral thesis and unreviewed preprint, and has not yet completely zeroed out the effects in all datasets subjected to it.

Flores’ thesis and subsequent preprint explicitly report that psychological variables in the matrix experiments exhibit high intercorrelations and that simulations incorporating this dependence reduce the apparent significance of earlier studies; this is quoted verbatim in MPI‑E009 and comes from within the same research program that had previously reported positive results. This concession against interest strongly supports the methodological critique that the founding analyses over‑counted significant cells by assuming independence. The main opposing consideration is that we do not yet have a fully independent, published, dependence‑aware reanalysis of all founding datasets quantifying “substantially diminish” in standardized effect‑size or p‑value terms, and Flores herself still finds some significant results under permutation tests, suggesting the critique weakens but may not fully erase the anomalies.

Flores, working inside the program under Watt and Walach, reports that psychological variables were highly intercorrelated and that simulations respecting that dependence diminished the initial significant results — a concession against interest, quoted in the 2023 preprint. That directly attacks the load-bearing statistic of MPI-C001. The strongest opposing consideration is that the full quantitative reanalysis of the 1986–2006 founding datasets has not been published as a standalone reviewed result, and Flores herself still reported three of five thesis studies significant under permutation tests, so the critique shrinks the origin story without zeroing out the later program on its own terms.

This is the sharpest technical attack on the program's foundation and it comes from inside: Flores's own thesis work found high intercorrelation among the psychological variables, and the 2023 successor states plainly that when dependence-respecting simulations were incorporated, "the results from these studies diminished" — a concession against interest that raises credibility. Two limitations hold it at provisional: the full dependence-aware reanalysis of the founding datasets has never been published as a standalone, independently reviewed result, so the magnitude of the shrinkage is summarized rather than demonstrated; and the critique has not zeroed the program even on its own terms, since the same project reports three of five studies significant under permutation analyses. The strongest opposing consideration is precisely that residual: the critique shrinks the founding results without eliminating structured excesses under methods designed to answer it.

The first conjunct is close to certain on general statistical grounds and is directly reported by Flores: behavioural variables derived from one key-pressing session are heavily intercorrelated, so counting significant cells against independence overstates the evidence. The second conjunct — that dependence-aware reanalyses substantially diminish the founding results — is thin. It rests on a single sentence in an unreviewed preprint whose referent ("results from these studies") plausibly denotes Flores's own replications rather than the 1986–2006 originals, whose raw data are unpublished and therefore not reanalysable. The strongest opposing consideration is internal to the same work: Flores's permutation analyses shrank but did not eliminate the effect, retaining significance in three of five studies. The direction of the evidence favours the claim; its quantity does not support high confidence.

Unchanged: the dependence critique comes from inside the program (MPI-E007), a concession against interest that raises credibility, but it has not yet appeared as a standalone peer-reviewed reanalysis of the founding datasets.

The dependence critique is a concession against interest from within the program (MPI-E007), which raises its credibility. But it has not appeared as a standalone peer-reviewed reanalysis of the founding datasets, so the 'substantially diminish' half remains unaudited.

The dependence critique originates inside the program (MPI-E007), a concession against interest that raises credibility. It has still not appeared as a standalone peer-reviewed reanalysis of the founding datasets, so the 'substantially diminish' half remains unaudited.

The dependence critique is a concession against interest from inside the program (MPI-E007), which raises its credibility considerably. But it has not appeared as a standalone peer-reviewed reanalysis of the founding datasets, so the 'substantially diminish' half remains unaudited.

Research originating from within the program itself demonstrates that psychological variables in these matrices are highly intercorrelated (MPI-E007). When permutation and Monte Carlo simulations properly account for this structural dependence, the significance of the original results diminishes notably. The strongest opposing consideration is that even under these dependence-aware permutation tests, partial statistical significance still survives in a subset of the modern replication studies.

Flores’s thesis and subsequent preprint explicitly state that the psychological variables in matrix experiments exhibit high intercorrelations and that simulations respecting this dependence reduce the apparent strength of earlier significant results (MPI-E007, MPI-E016). This directly supports the claim that independence assumptions were violated and that dependence-aware reanalysis diminishes the founding effects. The strongest contra point is that Flores herself still finds three of five thesis studies significant under permutation tests that account for dependence, implying that while the critique weakens the original claims, it does not clearly eliminate all evidence for an excess-correlation effect.

Flores, working inside the program under Watt and Walach, found high intercorrelations among the psychological variables and reported that incorporating simulations diminished the earlier significant results — a concession against interest, later restated in the 2023 preprint. That is direct evidence that the independence assumption of the founding analyses was false and that correcting it shrinks the headline surpluses. The strongest opposing consideration is that the full quantitative reanalysis of the 1986–2006 datasets has not been published as a standalone reviewed result, and Flores herself still reported three of five thesis studies significant under permutation tests that respect dependence, so the critique weakens the origin story without zeroing out the program on its own terms.

The first half of the claim is highly plausible and partly a priori: batteries of behavioural and psychological variables are routinely intercorrelated, so counting significant cells against an independence null overstates evidence — this leans on general statistical calibration as much as on the records here. The second half, that dependence-aware reanalysis substantially diminishes the founding results, rests on a single line of work (Flores 2021 thesis, summarized in a 2023 preprint), reported as a concession against interest but never published as a standalone, independently reviewed reanalysis, and apparently not applied to the 1986–2006 datasets themselves, whose raw data are unavailable. The strongest opposing consideration is internal: Flores's own permutation analyses still returned three of five significant, so the critique shrinks rather than zeroes the paradigm. Confidence is low because one unreviewed source is carrying a load-bearing technical verdict.

The dependence critique is a concession against interest from inside the program (MPI-E007), which raises credibility, but it has not appeared as a standalone peer-reviewed reanalysis of the founding datasets, so the 'substantially diminish' half remains unaudited.

The dependence critique is a concession against interest from inside the program (MPI-E007), which raises its credibility, but it has not been published as a standalone peer-reviewed reanalysis of the founding datasets, so the 'substantially diminish' half remains unaudited.

Reported from inside the research program itself (Flores worked under Watt and Walach), which raises its credibility as a concession against interest. The full quantitative reanalysis of the founding datasets has not been published as a standalone, independently reviewed result.