Pre-registered, falsifiable, and reported in full — including the nulls
A four-paper series traces one question end to end: the unexplained non-shared developmental residual (Foundation), a multi-axis common-cause model of prenatal hormone exposure and its falsifiable readability prediction (Theory), a pre-registered held-out test that refuted that prediction (The test), and a reflection on the bright line that separates this from physiognomy (Capstone) — accompanied by a standalone methods paper on label-leakage hygiene for self-typed corpora.
Twin studies reveal a robust and under-exploited pattern. Across thousands of human traits, roughly half of phenotypic variance is genetic. Shared family environment contributes little for most traits. A large remaining fraction is non-shared. This is individual-specific variance unexplained by either genes or household. We synthesize the personality and facial-morphology twin literatures. We argue that this non-shared residual is a plausible natural home for a prenatal developmental contribution. Conventionally, this residual is treated as post-natal idiosyncrasy or noise. We distinguish this residual sharply from the missing heritability gap. The two…
Prenatal androgen exposure acts as an accepted common cause of co-occurring physical and behavioral traits - digit ratio, spatial cognition, and the anatomical-behavioral profile of congenital adrenal hyperplasia all trace to a shared hormonal gradient. We generalize this single-axis precedent into a multi-axis common-cause model: across critical fetal windows, the intrauterine hormonal milieu (testosterone, cortisol, thyroid, estradiol, and others, varying in dose and timing) simultaneously shapes physical structure, including the face, and neural architecture, so that observable form constitutes a correlated trace of the developmental event that shaped disposition rather…
A multi-axis common-cause model of prenatal hormone exposure predicts that a trait-axis's readability from facial morphology increases monotonically with the strength of its prenatal hormonal organization. Operationalized on a community-typed celebrity facial dataset, the model forbids a specific effect-size ordering: a "function" axis reads at least as strongly as a sex-modality axis, and both out-read two weakly-organized "letter" axes (function ≥ sex-modality > letters). An exploratory discovery cohort (707 identities) appeared to support this - a function-axis effect of Cohen's d ≈ −0.47 surviving sex adjustment - but that analysis…
Physiognomy, the claim that a person's face causes or directly displays their character, is false, and modern machine-learning revivals of it have become a reputational and ethical landmine. Yet a structurally different claim hides in the same neighborhood: a single prenatal event (the intrauterine hormonal milieu acting across critical developmental windows) can shape both physical structure and neural architecture so that observable form becomes a correlated trace of a developmental signal rather than a cause of mind. This Perspective draws the bright line between the two. Physiognomy asserts body-to…
Researchers increasingly build supervised prediction sets from creator media: a person speaks on video, an automatic-speech-recognition (ASR) system transcribes them, and a human annotator who watched the same video assigns a label, which then serves as the target for the transcript-as-input. This common pipeline carries a silent, structural failure mode shared provenance. When the text channel and the label both descend from one recording, naive text to label prediction scores well for reasons unrelated to the construct: the model re-reads the cues the annotator read, or reads the construct's own…