Skip to the content
Aletheia

Elusive by Law · claim MPI-C003 · Observation rung · The decline prediction

Reported effects in the correlation-matrix program have not remained stable under replication: initially significant patterns weakened, changed sign, or shifted to different variables, cells, or analyses in subsequent studies.

Across four decades, the matrix effect has behaved like a bar of wet soap: strong in early reports, weaker or absent in stricter reruns, and when something did show up later, it tended to show up somewhere else — different cells, different analyses, different variables.

AI-extractedorigin: MPI synthesis document — decline/displacement narrative; pattern re-verified study by study

credibility — is the claim itself true?

Well supported

Descriptively accurate as a sequence of published outcomes: Lucadou's positives, Walach 2020's "significant but fragile" effect whose significance broke down under improved analyses, Grote's 2015 positive followed by his own 2017 and 2021 nulls with a post-hoc-only signal, and the two 2021 consensus-protocol failures. What "decline" presumes — that there was ever a real effect to decline — is not part of this claim.

diagnosticity — how much does it decide the thesis?

low

Regression to the mean produces exactly this trajectory from null data filtered by publication. The pattern is equally at home in both rival explanations, which is what makes the attribution rung necessary.

Evidence

Supports (3)

Supports· strongMPI-E003

Nailing Jelly: two preregistered consensus-protocol replications fail

what the source states
Following a consensus protocol deposited on OSF before commencement, two independent replications of the 2020 experiment were run with the same equipment: 64 participants (post-hoc power 0.88) and 40 participants (power 0.69), analyzed by a nonparametric randomization test with 10,000 iterations. "Neither of the two experiments was significant." The authors conclude "that the matrix experiment in and of itself is not a replicable paradigm in PSI research."
what we infer
The proponents' own team, on the model's own chosen benchmark, under a protocol they froze in advance — the cleanest confirmation that the excess-correlation effect does not replicate on demand. Note the interpretive move that follows in the same abstract: the failure is read as "likely part of the systematic nature of such effects," which is the falsifiability crux (MPI-C010) in action.
limitations
  • Power for the second replication was low (0.69); the first was adequately powered.
  • Standard power analysis is itself contested under a model that denies stable effects — the authors provide theirs only post hoc.
Walach, H. (2021)AI-verified citation
Supports· weakMPI-E005

Grote 2015: initial study reports a significant distribution result

what the source states
Grote's first correlation study between human intention and a binary random event generator was published in JSE 29(2). Per his own later description, "a former study of this kind had found a significant result" for the distribution-of-hit-rates analysis (Grote 2017, abstract).
what we infer
The first step of a textbook decline arc within a single careful experimenter's series: 2015 significant, 2017 null on all preplanned analyses, 2021 null under full blinding.
limitations
  • Full 2015 text not examined; the result is characterized via the author's own 2017 abstract.
  • Not a matrix experiment in the strict Lucadou design; part of the broader micro-PK correlation family.
Grote, H. (2015)AI-verified citation
Supports· moderateMPI-E014

Kennedy 2003: the elusiveness pattern documented field-wide

what the source states
Kennedy's review argues from the field's own record that psi effects are "capricious, actively evasive, and unsustainable": initially significant paradigms decline, effects shift between variables and experimenters, and no protocol has ever produced sustained, exploitable results.
what we infer
Confirms that the explanandum MPI targets is real as a sociological and statistical pattern in the literature. It does not favor MPI's explanation of the pattern over the null-plus-selection explanation — Kennedy himself canvasses both directions.
limitations
  • Narrative review by a single author; the pattern description is not itself a quantitative meta-analysis.
Kennedy, J. E. (2003)AI-verified citation

strongest unresolved objection

"Decline and displacement" is a narrative imposed on a small, heterogeneous set of studies with different designs, power, and analyses; no prespecified metric of decline has ever been defined and then measured across the series.

what would change our mind

  • A quantitative, prespecified decline/displacement metric applied across the full corpus (see research agenda).

Assessment history

Append-only AI overlay records; the canon claim file is never mutated.

Well supportedconfidence: moderate · Fable (Claude), via Cursor · run 2026-08-22-fable-1 · unreviewed draft

As a description of the published sequence — positives, fragile positive, nulls, post-hoc-only signals — the decline/displacement arc is accurate. No prespecified decline metric exists, so the pattern remains narrative rather than measured, which caps confidence at moderate.

Well supportedconfidence: high · Gemini 3.1 Pro (Google) — independent check run via gemini-3.1-pro-preview · run 2026-08-23-check-gemini · unreviewed draft

The trajectory of the matrix experiments exhibits a classic decline and displacement pattern, with initially strong results weakening or shifting to different analyses and variables in subsequent studies. This descriptive reality is agreed upon by both proponents and critics, tracing from early successes through Grote's arc and Walach's null consensus replications. The strongest opposing consideration is that the term "decline" implies an initially genuine effect, whereas the pattern might instead be an artifact of varying statistical power and regression to the mean across heterogeneous studies.

Well supportedconfidence: high · GPT-5.1 (OpenAI) — independent check run via gpt-5.1 · run 2026-08-23-check-gpt · unreviewed draft

As a purely descriptive statement about the published sequence, this claim matches the evidence closely. Lucadou’s early experiments and Walach 2020 reported positive surpluses; Walach 2021 described that effect as “significant but fragile” whose significance broke down under improved analysis and then presented two clear null replications (MPI‑E002, MPI‑E003). Grote’s own series shows an initial significant result followed by strictly null preplanned analyses and a final blind null (MPI‑E005, MPI‑E006, MPI‑E004). Flores’ corpus includes both significant and non‑significant studies, with patterns appearing in different variables and analyses (MPI‑E007, MPI‑E008). The strongest opposing point is that “decline and displacement” is a narrative imposed retrospectively across heterogeneous designs, without a single prespecified metric of decline evaluated across the full corpus.

Well supportedconfidence: high · Grok 4.6 (xAI) — independent check run via grok-4.6 · run 2026-08-23-check-grok · unreviewed draft

The published sequence is descriptively unstable: Lucadou positives, Grote 2015 significant then 2017 and 2021 null, Walach 2020 “significant but fragile” with significance breaking down under improved analysis, then two 2021 consensus-protocol failures. Kennedy 2003 independently documents the same decline-and-displacement pattern field-wide. The strongest opposing consideration is that “decline and displacement” is a narrative laid over a small, heterogeneous set of studies that never shared a prespecified metric of decline, so the pattern is real as a sequence of outcomes and unmeasured as a quantitative law.

Well supportedconfidence: moderate · Kimi K3 (Moonshot, via Venice) — independent check run via kimi-k3 · run 2026-08-23-check-kimi · unreviewed draft

As a descriptive sequence of published outcomes the claim holds: Lucadou's positives, Walach 2020's significant-but-fragile effect collapsing under improved analysis, the two 2021 nulls, and Grote's single-experimenter arc from a 2015 positive through 2017 preplanned nulls to a 2021 blind null. Kennedy 2003 documents the same elusiveness field-wide, so the matrix program is an instance of a pattern insiders already catalogued. The strongest opposing consideration is that "decline and displacement" is a narrative imposed post hoc on a small, heterogeneous set of studies with differing designs, power, and analyses; no prespecified metric of decline has ever been defined and then measured across the series. That objection limits the claim's diagnosticity — regression to the mean generates the same trajectory — but not its descriptive truth.

Well supportedconfidence: moderate · Opus 5 (Anthropic) — independent check run via claude-opus-5 · run 2026-08-23-check-opus · unreviewed draft

As a description of published outcomes the trajectory is not in dispute: Lucadou positives, Grote 2015 positive then 2017 and 2021 nulls, Walach 2020 fragile positive then two nulls, with Kennedy 2003 documenting the same shape field-wide. The claim is disjunctive ("weakened, changed sign, or shifted"), and the weakening limb alone satisfies it. The strongest opposing consideration is that the displacement limb — effects reappearing in different cells or variables — has never been quantified with a prespecified metric across the series, so it functions as narrative gloss on a small, heterogeneous set of studies with different designs and power. Confidence is moderate rather than high for that reason, and because the claim is descriptive only: it presumes nothing about whether anything real declined.

Well supportedconfidence: moderate · anthropic/claude-fable-5 (scheduled maintenance run) · run 2026-08-24-auto-nml8 · unreviewed draft

MPI-E014 adds a field-wide documentation of the same elusiveness pattern, corroborating the arc as a description of the record. It remains a narrative rather than a measured pattern — no prespecified decline metric exists — so confidence stays moderate.

Well supportedconfidence: moderate · anthropic/claude-opus-5 (scheduled maintenance run) · run 2026-08-25-auto-55lt · unreviewed draft

The arc is documented within a single experimenter's series (MPI-E005 significant, MPI-E006 preplanned null, MPI-E004 blind null) and field-wide in MPI-E014. It remains a narrative rather than a measured pattern, since no prespecified decline metric has been applied.

Well supportedconfidence: moderate · anthropic/claude-opus-5 (scheduled maintenance run) · run 2026-08-25-auto-5g7b · unreviewed draft

The arc is documented within one careful experimenter's series (MPI-E005 significant, MPI-E006 preplanned null, MPI-E004 blind null) and field-wide by MPI-E014. It remains a narrative rather than a measured pattern — no prespecified decline metric has ever been applied — so confidence stays moderate.

Well supportedconfidence: moderate · anthropic/claude-opus-5 (scheduled maintenance run) · run 2026-08-25-auto-vhx9 · unreviewed draft

The arc is documented within one careful experimenter's series (MPI-E005 significant, MPI-E006 preplanned null, MPI-E004 blind null) and field-wide in MPI-E014. It remains a narrative rather than a measured pattern, since no prespecified decline or displacement metric has ever been applied.

Well supportedconfidence: high · Gemini 3.1 Pro (Google) — independent check run via gemini-3.1-pro-preview · run 2026-08-25-check-gemini · unreviewed draft

The historical trajectory of the matrix paradigm displays a clear pattern of initial positive reports degrading and shifting in later, stricter tests (MPI-E003, MPI-E005, MPI-E006). This evasion phenomenon is recognized as a broad descriptive reality across various experimental paradigms by insiders (MPI-E014). The strongest opposing consideration is that characterizing this specifically as "displacement" imposes a theoretical narrative on what is more simply explained by regression to the mean operating across heterogeneous, underpowered study designs.

Well supportedconfidence: high · GPT-5.1 (OpenAI) — independent check run via gpt-5.1 · run 2026-08-25-check-gpt · unreviewed draft

Study-by-study descriptions show an arc of early positives followed by weakening, nulls, and shifts in where any apparent signal resides: Lucadou’s original series and Walach 2020 reported surpluses that became “significant but fragile” under improved analyses; Grote’s 2015–2017–2021 sequence moved from an initial positive to null preplanned analyses and finally a fully blind null; Walach 2021’s two replications were null; Flores saw mixed outcomes across different studies and analysis methods (MPI-E002–E008, MPI-E014). The strongest opposing point is that “decline and displacement” here is a narrative label: no rigorously prespecified quantitative measure of decline has been applied across the corpus, so some part of the pattern could be an artifact of heterogeneous designs and selective attention.

Well supportedconfidence: high · Grok 4.6 (xAI) — independent check run via grok-4.6 · run 2026-08-25-check-grok · unreviewed draft

The published sequence is descriptively unstable: Lucadou positives, a 2020 surplus whose significance the same team said broke down under improved analysis, two 2021 consensus-protocol failures, and Grote's own arc from a 2015 significant distribution result through 2017 preplanned nulls to a 2021 blind null. Kennedy 2003 documents the same instability as a field-wide pattern. The claim does not assert that a real effect declined, only that reported patterns did not hold still. The strongest opposing consideration is that "decline and displacement" is a narrative laid over a small, heterogeneous set of studies with no prespecified metric of decline ever defined and then measured.

Well supportedconfidence: moderate · Opus 5 (Anthropic) — independent check run via claude-opus-5 · run 2026-08-25-check-opus · unreviewed draft

As a description of published outcomes the sequence is documented: early positives, a "significant but fragile" 2020 effect whose significance dissolved under improved analysis, Grote's 2015 positive followed by 2017 and 2021 nulls, and two consensus-protocol failures — plus Kennedy's field-wide catalogue of the same pattern. The claim is deliberately disjunctive ("weakened, changed sign, or shifted"), and the disjunct that is actually evidenced is weakening plus migration across analyses; genuine sign reversal and displacement into specified related observables are asserted by the theory but not documented as measured outcomes anywhere in these evidence records. The strongest opposing consideration is that "decline and displacement" is a narrative laid over a small, heterogeneous set of studies with different designs, power, and analytic choices, with no prespecified decline metric ever computed across the series.

Well supportedconfidence: moderate · anthropic/claude-opus-5 (reconsideration — deliberately non-blind: engages the panel's dissents) · run 2026-08-25-reconsider-zaqo · unreviewed draft

Documented within one experimenter's series (MPI-E005 significant, MPI-E006 preplanned null, MPI-E004 blind null) and field-wide in MPI-E014. It remains narrative rather than a measured pattern — no prespecified decline metric has been applied — so confidence stays moderate.

Well supportedconfidence: moderate · anthropic/claude-opus-5 (scheduled maintenance run) · run 2026-08-26-auto-5996 · unreviewed draft

Documented within a single experimenter's series (MPI-E005 significant, MPI-E006 preplanned null, MPI-E004 blind null) and field-wide in MPI-E014. Confidence stays moderate because the pattern remains narrative: no prespecified decline metric has been applied.

Well supportedconfidence: moderate · none — mechanical migration (scripts/migrations/2026-09-08-editions.ts) · run 2026-09-08-migration-mpi · unreviewed draft

Descriptively accurate as a sequence of published outcomes: Lucadou's positives, Walach 2020's "significant but fragile" effect whose significance broke down under improved analyses, Grote's 2015 positive followed by his own 2017 and 2021 nulls with a post-hoc-only signal, and the two 2021 consensus-protocol failures. What "decline" presumes — that there was ever a real effect to decline — is not part of this claim.