---
title: "Evidential weight rests on independence — and its absence, appearing as excessive perfection, is itself a signature of defect"
type: "observation"
status: "seedling"
audit_status: "seedling — a synthesis across three constituent claims, each individually seedling and each capture-verified only (not independently re-fetched this promotion): [[claim-fisher-1936-flagged-mendels-pea-data-as-improbably-close-fit]], [[claim-sanhedrin-unanimous-guilty-verdict-acquits-the-defendant]], [[claim-simonsohn-fabrication-flagged-via-excessive-similarity-to-random-sampling]]. The observation is a structural reading across those three plus two pre-existing vault notes, not a new external fact; it cannot outrank its weakest supporting leg (the Tier-3 Sanhedrin note). See [[question-verify-suspicious-perfection-hop-primaries]]."
source_url: "https://blog.thegrandlocus.com/2016/04/did-mendel-fake-his-results; https://www.sefaria.org/Mishneh_Torah,_The_Sanhedrin_and_the_Penalties_within_Their_Jurisdiction.9.1; https://bitss.org/just-post-it-the-lesson-from-two-cases-of-fabricated-data-detected-by-statistics-alone-by-uri-simonsohn"
source_author: "Synthesis across Fisher (1936, via The Grand Locus), Maimonides (via Sefaria/Pace), and Uri Simonsohn (Data Colada/bitss.org)"
source_date: "2026-07-11T00:00:00.000Z"
source_venue: "See constituent claim-notes for individual venues"
source_quote: "it remains a possibility among others that Mendel was deceived by some assistant who knew too well what was expected"
source_tier: 3
provenance: "Promotion from 10-inbox/raw/2026-07-11-hop-suspicious-perfection.md, 2026-07-12"
origin: "batch"
derived_from: "10-inbox/raw/2026-07-11-hop-suspicious-perfection.md"
date_created: "2026-07-12T00:00:00.000Z"
writer_model: "claude-sonnet-5"
tags: ["epistemology","research-integrity","fraud-detection","legal-history","statistics","cross-domain-bridge","independence"]
drafted_in: ["2026-07-13-enlightenment-backwards","2026-07-13-the-noise-is-the-evidence","enlightenment-backwards","the-noise-is-the-evidence"]
---


A single epistemic law recurs across law and statistics, over roughly two millennia: real evidence carries the noise of independent origin, so *too little* noise — too-perfect agreement, too-close a fit — is not confirmation but a warning sign that something in the evidentiary chain is not actually independent.

Three anchors triangulate it. [[claim-fisher-1936-flagged-mendels-pea-data-as-improbably-close-fit|R. A. Fisher (1936)]] found Mendel's pea-ratio data fit the predicted 3:1 split so closely (P≈0.99993) that he floated an over-eager assistant rather than accept it as ordinary sampling luck — the founding "fraud-by-overfit" case, genuinely contested to this day. [[claim-sanhedrin-unanimous-guilty-verdict-acquits-the-defendant|Talmudic law]] (Maimonides, *Mishneh Torah*, Sanhedrin 9:1) codifies the same instinct in reverse: a unanimous guilty verdict in a capital case is disqualifying, because total agreement among judges means no one argued the required defense. [[claim-simonsohn-fabrication-flagged-via-excessive-similarity-to-random-sampling|Uri Simonsohn's fabrication forensics]] formalizes it statistically: fabricated data is caught when summary statistics show "excessive similarity ... inconsistent with random sampling."

The vault held two unlinked notes that turn out to be opposite faces of this same law rather than a loose thematic resemblance. [[claim-gersick-1991-deep-structure-quotes-confirmed-verbatim]] grades a *true* quotation down to a "half-proof" because its only carrier is a single, non-independent mirror — the medieval *système de preuve légale* discounting evidence for want of independent corroboration. [[claim-hirsch-forensics-drove-dias-superconductivity-retraction]] is the mirror image: a *false* result exposed because its magnetic-susceptibility data was implausibly smooth for genuine lab noise. One is positive verification (a true claim penalized for thin corroboration), the other negative detection (a false claim caught for too much internal agreement) — but both cash out the same rule: independence is what evidence is made of, and its absence is legible.

A related but distinct failure mode is worth flagging rather than folding in: [[claim-bell-burnell-1967-pulsar-scruff-first-dismissed-as-interference|Jocelyn Bell Burnell's 1967 pulsar signal]] was initially dismissed as noise precisely *because* it was too regular to be believed — the opposite miscalibration, where excess regularity led to wrongly discounting a real signal rather than wrongly trusting a fake one. Both errors run on the same expectation ("real data is supposed to look messy"), but one overreads perfection as fraud and the other underreads it as artifact; they are not the same claim and should not be merged.

> [!note] Seek's commentary:
> The seed question was whether two 0.75-cosine, unlinked notes "genuinely connect." They do, but not as a pair — as two points on a much older curve. Naming the curve is the actual find: the vault's own tiering apparatus (grading a verbatim quote down for thin corroboration) and modern data forensics (grading a dataset down for too little noise) are doing the identical epistemic move, one just runs it forward against fabrication and the other runs it backward against source thinness. All three new anchors here rest on single, un-re-fetched secondary or tertiary sources — I kept this whole cluster at `seedling` on purpose; it's a good story, not yet a hardened one. — Seek