Regression to the mean, analytic flexibility, publication bias, and statistical dependence among matrix variables are jointly sufficient to explain the correlation-matrix record — the early positives and the decline — without any anomalous correlation.

Every component mechanism is documented in or beside this corpus, and the blind-analysis null is exactly its predicted outcome. It stops short of well_supported because no calibrated simulation has yet reproduced the specific published record, and the best analog effort (Bierman 2016) explicitly failed to cover the full distance in its own database.

Every component of the conventional explanation—statistical dependence, analytic flexibility, coding errors, and publication bias—has been empirically documented within this specific literature or its direct antecedents. It offers a highly parsimonious explanation for the entirety of the observed decline arc. The strongest opposing consideration is that a comprehensive, quantitative demonstration showing these mechanisms account for 100% of the founding matrix successes has not yet been published, and some positive results survive dependence-aware permutation tests.

Each component of the conventional explanation is independently evidenced. Bösch’s meta‑analysis finds that REG micro‑PK as a whole is best explained by publication bias and small‑study effects (MPI‑E015); Bierman’s simulations show that QRPs at realistic rates can explain a large fraction of reported psi‑like effects (MPI‑E016); Maier & Dechamps demonstrate a concrete case where a software error produced a spurious micro‑PK correlation (MPI‑E017); and Flores’ Monte‑Carlo work identifies high intercorrelations among matrix variables and shows that dependence‑aware simulations diminish the original surpluses (MPI‑E007, MPI‑E009). Grote’s 2017 study provides a direct example of post‑hoc analysis choices extracting a significant result from data where all preplanned tests are null (MPI‑E006). Together, these make it plausible that the matrix record could arise from null data subject to these influences. The strongest opposing consideration is that no single quantitative model has yet been fit to the actual matrix corpus to demonstrate sufficiency — and some preregistered or dependence‑aware positives (Walach 2020; parts of Flores’ work) remain unexplained in detail, leaving a residual ambiguity.

Each named mechanism is attested in or next to this corpus: dependence inflating cell counts (Flores), analytic flexibility manufacturing significance (Grote 2017’s post-hoc p = 0.012 against three preplanned nulls), a coding error producing a false micro-PK positive (Maier and Dechamps), and publication-bias sensitivity in the parent REG literature (Bösch 2006). Joint sufficiency for this specific matrix record has not been shown; Bierman’s neighboring ganzfeld simulation explained about sixty percent of that effect, not all of it, and only under realistic QRP rates. The strongest opposing consideration is that Walach 2020 was preregistered-era, adequately powered, and still reported a surplus, and Flores found effects in three of five studies using analyses designed to respect dependence — the mundane account predicts a cleaner kill than the file contains.

Each component mechanism is documented in or immediately adjacent to this corpus: violated independence assumptions (Flores's Monte Carlo work), analytic flexibility demonstrated live in Grote 2017's post-hoc-only significance, a coding error manufacturing a celebrated micro-PK effect in the neighboring line, and publication-bias sensitivity in the parent REG literature. The strongest opposing consideration has two prongs: the sufficiency claim is quantitative and has never been demonstrated against this specific corpus — no simulation calibrated to the matrix record has reproduced it — and the best-controlled positive (Walach 2020, preregistered and powered) is exactly the kind of result this account predicts should not occur, with Flores's permutation-surviving results beside it. Bierman's QRP simulation cuts both ways: it quantifies how far mundane mechanisms really go, and in the best-studied neighboring database they covered about sixty percent, not everything. Plausible, mechanism-rich, but not yet shown sufficient.

Every component mechanism is documented operating inside or adjacent to this literature: dependence inflating cell counts (Flores), analytic flexibility manufacturing p = 0.012 from null data (Grote 2017), a coding error manufacturing a micro-PK effect (Maier and Dechamps), and publication bias as the most economical account of the parent REG meta-analysis (Bösch 2006). The account also correctly predicted the outcome of the corpus's strictest test. What holds it short of well_supported is that "jointly sufficient" has never been demonstrated by a simulation calibrated to the actual matrix record. The strongest opposing consideration is the residue: a preregistered, adequately powered 2020 replication reported a surplus anyway, and Flores's permutation analyses — designed to kill exactly this artifact — left three of five studies significant.

Unchanged: each component mechanism is documented in or beside this corpus (MPI-E006, MPI-E015, MPI-E017), and the blind-analysis null is its predicted outcome. It stops short of well_supported because no calibrated simulation reproduces this specific record, and MPI-E016 shows the analog effort covered only ~60% of a neighboring effect.

Each component is documented nearby — analytic flexibility (MPI-E006), publication bias (MPI-E015), coding error (MPI-E017), dependence (MPI-E007) — and MPI-E004 is the outcome it predicts. It falls short of well_supported because no calibrated simulation reproduces this specific record and MPI-E016 shows the analogous exercise covered only ~60% of a neighboring effect.

Each component mechanism is documented nearby — analytic flexibility (MPI-E006), publication bias (MPI-E015), coding error (MPI-E017), dependence (MPI-E007) — and the blind-analysis null (MPI-E004) is its predicted outcome. It falls short of well_supported because no calibrated simulation reproduces this specific record, and MPI-E016 shows the analog effort covered only ~60% of a neighboring effect.

Each component is documented nearby — analytic flexibility (MPI-E006), publication bias (MPI-E015), coding error (MPI-E017), dependence (MPI-E007) — and MPI-E004 is its predicted outcome. Short of well_supported because no calibrated simulation reproduces this specific record and MPI-E016 shows the analogous effort covering only ~60% of a neighboring effect.

Documented instances of analytic flexibility (MPI-E006), software errors producing false positives (MPI-E017), and statistical dependence inflating significance (MPI-E007) establish that these mundane mechanisms actively govern this literature. Simulations also show questionable practices explain the vast majority of related flagship phenomena (MPI-E016). The strongest opposing consideration is that a specific, quantitative simulation fully reproducing the exact decline and displacement record of the matrix corpus has not yet been formally calibrated and published.

Each component of the conventional explanation is concretely instantiated here: dependence among psychological variables and its inflationary effect on significant-cell counts are documented by Flores’s Monte Carlo and permutation analyses (MPI-E007, MPI-C009); analytic flexibility is directly illustrated by Grote’s 2017 post-hoc-only positive (MPI-E006); publication bias and small-study effects in the broader REG literature are highlighted by Bösch 2006 (MPI-E015); and a major micro-PK effect’s collapse under a coding error (Maier & Dechamps 2022) shows how false positives can arise and persist (MPI-E017). Together, these make it provisionally credible that such mechanisms suffice to explain the matrix record, though no single quantitative model has yet reproduced the entire corpus, and some larger, more conservative positives (e.g., Walach 2020) remain as a counterweight.

Dependence-inflated cell counts, a post-hoc rescue of a null study (Grote 2017, p = 0.012 after three preplanned misses), a coding-error false positive in a neighboring micro-PK line, and a large REG meta-analysis that favors publication bias are all in this corpus or its immediate parent literature, and they jointly predict early selected positives followed by decline under tighter methods — which is the record we have. Sufficiency for this specific matrix series has not been demonstrated by a calibrated simulation, and Bierman 2016 covered only about sixty percent of a different flagship effect, so the claim stops short of well_supported. The strongest opposing consideration is that Walach 2020 and parts of Flores survived procedures intended to block the very artifacts invoked, which the mundane account says should not happen cleanly.

Each component is independently evidenced within or adjacent to this corpus: dependence inflating cell counts (Flores), analytic flexibility manufacturing significance from the same null data (Grote 2017's post-hoc p = 0.012 against three preplanned nulls), an ordinary coding error manufacturing a micro-PK effect (Maier and Dechamps), and publication bias as the most economical account of the parent REG literature (Bösch 2006). The account also made and won a forward prediction: the strictest design returned null. What blocks a stronger verdict is that "jointly sufficient" has never been demonstrated quantitatively against this specific corpus, and the closest calibration — Bierman's QRP simulation on ganzfeld — covered about 60% of that effect while requiring implausible parameters to reach 100%. The strongest opposing consideration is the residue: a preregistration-era two-lab positive and three of five permutation-based positives that this account must attribute to selection and flexibility without having shown it can.

Every component is documented nearby — analytic flexibility (MPI-E006), publication bias (MPI-E015), coding error (MPI-E017), dependence (MPI-E007) — and MPI-E004 is its predicted outcome. Short of well_supported because no calibrated simulation reproduces this specific record and MPI-E016 shows the analog effort covered only ~60% of a neighboring effect. Gemini's and GPT's caution here is conceded and already priced in.

Each component is documented nearby — analytic flexibility (MPI-E006), publication bias (MPI-E015), coding error (MPI-E017), dependence (MPI-E007) — and MPI-E004 is its predicted outcome. Short of well_supported because no calibrated simulation reproduces this specific record and MPI-E016 shows the analogous effort covered only ~60% of a neighboring effect.

Each component mechanism is documented in or near this corpus: the dependence problem (Flores's Monte Carlo work), analytic flexibility (Grote 2017's post-hoc-only significance), error-driven false positives (Maier & Dechamps), publication-bias sensitivity (Bösch 2006). What keeps this short of well_supported: no one has yet shown quantitatively that these mechanisms reproduce the specific published record, and the strongest QRP simulation in the neighborhood (Bierman 2016, on ganzfeld data) explained about 60% of that effect, not all of it.