Xenodex Sciences

An independent research program in developmental-trace inference

The Question

The program investigates a single empirical question: does facial morphology carry a measurable developmental trace — a residue of the prenatal hormonal milieu — that correlates with psychological type?

The hypothesis under test treats the face and disposition as two downstream readouts of a shared upstream cause. If a common prenatal influence shapes both craniofacial development and aspects of temperament, then facial features could, in principle, carry partial information about type without any causal link running from one to the other.

This frames a common-cause question, not a physiognomic claim. The program does not assert that the face determines character, nor that character can be inferred by inspection. It asks only whether a shared developmental origin leaves a faint, statistically detectable correlation — and treats that as a falsifiable proposition.

A common-cause hypothesis, stated as a falsifiable prediction — not physiognomy.

How We Work

The method stance is the substance of the program. Each directional prediction is pre-registered before the confirmatory data are examined, so the analysis cannot be reshaped to fit the result. Confirmatory tests run on identity-disjoint held-out samples, where no person appearing in development data appears in evaluation.

Demographic confounds are treated as the primary threat — sex above all, because it correlates with both facial morphology and many trait measures and can manufacture an apparent signal that has nothing to do with development. Controls are specified in advance and reported whether or not they survive.

Null results are reported in full. When a pre-registered prediction fails, the failure is published with the same prominence the confirmation would have received.

Pre-registration, identity-disjoint held-out tests, confound control, and full reporting of nulls.

What We Found

The central prediction did not hold. The program forecast a facial "readability hierarchy" — an ordering of which type axes are more legible from the face. On pre-registered held-out data this prediction was refuted (n=83), and the refutation replicated at larger scale (n=549).

One signal survived: a sex-modality axis with a small effect (d ≈ 0.39). On inspection, the most plausible account is mundane — the axis most likely reduces to a sex-correlated demographic shortcut rather than a developmental trace. That is the classic physiognomy failure mode, and the program treats it as a confound to rule out, not a discovery to claim.

Reporting this null under pre-registration is the contribution. A refuted prediction, documented honestly, constrains the space of viable hypotheses in a way an unfalsifiable success story cannot.

The hierarchy was refuted on held-out data; the lone surviving signal most plausibly reflects a sex confound.

Publications

The work is documented as a connected series. A four-paper developmental-trace sequence moves from a meta-analytic motivation through a formal model to empirical, pre-registered tests — including the held-out refutation above. A standalone methods paper sets out the inference and evaluation protocol the series relies on.

The publications page lists each paper with its status and scope.

A four-paper developmental-trace series plus a standalone methods paper.

Contact

Michael Abdo, independent researcher. Correspondence on the methods, the data, or critiques of the findings is welcome.