~/shanegraffiti.com/research/genomic-bayesian-anchor

Shane Graffiti Inc. Semantic Adversarial Research Division 2026

IS IT YOU
OR YOUR
ENVIRONMENT?

Continuous physiological monitoring throws off a number every few seconds, but a number means nothing without a reference. Population norms can say whether a reading is typical for a group; they can't say whether it's typical for this person. A personal behavioral baseline is far more informative but it doesn't exist on day one. That's the cold-start problem. This framework proposes a third reference that exists before any behavioral data are ever collected: an exogenous genetic anchor, built from GWAS effect sizes applied to an individual's own genotype. Because genotype is fixed at conception, the anchor is immune to reverse causation, letting an observed reading decompose into a constitutional set point and an environmental deviation the one part of the signal that is candidate-causal and actionable. Six physiological domains are mapped end to end, from strongly replicated anchors (FTO, FADS1/2, FKBP5) down to a cautionary tier of contested candidate genes (SLC6A4, MAOA, DRD2), with explicit constraints for honest deployment: evidence-graded priors, dynamic decay toward a behavioral baseline, ancestry-matched effect sizes, and attribution rather than diagnosis.

Division Semantic Adversarial Research

Domain Causal AI / Personalized Physiology

Published arXiv 2606.13556 2026

Key Result Day-zero personalization, no behavioral data required

§ 2.0 Core Concepts

Two reference systems already exist for physiological interpretation, and both fall short. Population norms answer whether a reading is typical at the group level; a personal baseline is more informative but requires weeks of data to stabilize. The exogenous genetic anchor is a third reference, available from the first measurement, that sits between them.

Exogenous Genetic Anchor

A genomic estimate of an individual's constitutional set point, computed from GWAS effect sizes applied to the person's own genotype: Ĝ = μ + Σ βᵢgᵢ. Exogenous because genotype is fixed at conception the same property that licenses Mendelian randomization as a causal-inference tool.

Cold-Start Problem

A personal behavioral baseline needs roughly 7–30 days of consistent data per signal to stabilize. Until then, population norms can't separate a constitutionally high reading from an environmentally elevated one every early misattribution can seed incorrect causal hypotheses that outlast the baseline itself.

Causal Decomposition

Each observed reading P decomposes as δ = P − Ĝ. Because Ĝ is exogenous, δ is by construction non-genetic it isolates the portion of the signal that environment, behavior, and state can actually modify, narrowing the causal search space before any reasoning begins.

Uncertainty-Gated Inference

Ĝ is modeled as N(μ_G, σ²_G), not a fixed value. Evidence grade propagates directly into σ²_G strong anchors like FTO carry tight intervals, contested ones like DRD2 carry wide ones so a single noisy reading against a weak prior never licenses a confident attribution.

Prior-Decay Architecture

Ĝ_t = w(t)·Ĝ_genomic + [1−w(t)]·P̄_t. The genetic anchor carries full weight at day zero and decays toward a non-zero floor as a personal behavioral baseline accumulates nature and nurture trade off the interpretive role over time rather than competing for it once.

"Normal for Whom" Reversal

Two people can post an identical HRV of 55ms that falls comfortably inside the population band yet one is constitutionally suppressed below their own 80ms set point while the other is environmentally supported above their 30ms set point. Same number, opposite causal story.

§ 3.0 The Causal Framework

Genotype G sits at the root of the causal graph with no incoming edges nothing downstream can alter germline sequence. Environment E and G jointly produce the phenotype P, and decomposing the observed signal into a genetic set point and a deviation isolates the one component that behavior and environment can still move.

Nature Fixed at Conception

Genetic
Set Point

Ĝ is estimated from GWAS effect sizes and the individual's allele counts. It is not a prediction of the person's value it's a probabilistic prior, calibrated by evidence strength and carried with uncertainty proportional to that strength.

⇄ decompose

Nurture Everything Else

Non-Genetic
Deviation

δ = P − Ĝ reflects environment, lifestyle, disease state, and noise everything not fixed by genetics. It is the candidate-causal, actionable signal, and the only part of the reading a behavioral intervention can move.

The decomposition Ĝ = μ + Σᵢ βᵢ gᵢ // μ = population mean, βᵢ = GWAS effect size, gᵢ ∈ {0,1,2} = risk-allele count Ĝ ~ N(μ_G, σ²_G) // σ²_G grows with sampling error, unobserved heritability, and ancestry mismatch δ = P − Ĝ ~ N(P − μ_G, σ²_G + σ²_ε) // a deviation is informative only when |δ| substantially exceeds σ_G Ĝ_t = w(t)·Ĝ_genomic + [1 − w(t)]·P̄_t // w(t) → 1 at cold-start, w(t) → w_min > 0 as a personal baseline matures

§ 4.0 Six Physiological Domains

The decomposition is only as good as Ĝ, and Ĝ's quality varies enormously by domain. Strongest anchors first metabolic and fatty-acid signals carry real, mechanistically validated weight; autonomic and dopaminergic signals carry almost none.

Metabolic / Appetite

FTO rs9939609 shifts the satiety set point via cis-regulatory effects on IRX3/IRX5 in hypothalamic appetite circuits. Per-allele effect ≈0.36 kg/m² among the largest single-variant effects known for any complex trait.

Strong

Fatty-Acid / Inflammatory

FADS1/2 governs desaturase efficiency converting dietary fats into long-chain PUFAs. Low-efficiency variants shift constitutive inflammatory tone independent of diet two people on identical diets, different genetic baselines.

Strong

Stress-Axis / Cortisol

FKBP5 rs1360780 slows cortisol-feedback termination. Early-life adversity triggers allele-specific demethylation in risk carriers genotype alone doesn't fix the set point; epigenetic history is part of the anchor.

Moderate–Strong

Autonomic Tone (HRV)

Polygenic across dozens of sinoatrial-node loci, explaining only 0.9–2.6% of variance despite 47–64% twin heritability. Real but too diffuse to anchor confidently its value is illustrating the reversal, not predicting a number.

Moderate, Polygenic

Circadian / Chronotype

351 loci across PER1/2/3, CRY1, ARNTL identified in 697,828 individuals. SNP heritability (13.7%) sits well below twin heritability (40–50%) the missing-heritability gap, visible even in a well-powered study.

Moderate, Polygenic

Dopaminergic / Serotonergic

COMT, DRD2, SLC6A4 the most heavily marketed genes in consumer genomics, and the most consistently unreplicated. Belongs in a cautionary tier rather than as a set-point anchor.

Weak / Contested

§ 5.0 The Candidate-Gene Crisis

For decades, behavioral genomics chased named "candidate genes" selected for mechanistic plausibility. A pre-registered analysis of 18 historically prominent depression-gene hypotheses found no support across multiple large samples. The most celebrated gene-by-environment finding in psychiatry 5-HTTLPR × stress disappeared on aggregation. These genes stay in the framework only as a cautionary tier.

The cautionary candidate-gene tier claimed effect vs. replication status

Gene (Variant)	Claimed Association	Replication Status
COMT (Val158Met)	Prefrontal dopamine shaping executive function & stress response	Enzyme effect robust; behavioral effects small & reverse with age
SLC6A4 (5-HTTLPR)	S-allele raising anxiety/depression under stress	No robust main effect or G×E in large meta-analyses
MAOA (uVNTR)	Low-activity "warrior gene" linked to aggression	Small, inconsistent; documented forensic misuse history
DRD2 (TaqIA)	A1 allele, "reward deficiency" / addiction risk	Density effect debated; modest, contested in meta-analysis
DRD4 (7R VNTR)	7R allele raising novelty-seeking & ADHD risk	Mixed; novelty-seeking effect small, ADHD highly polygenic
DRD3 (Ser9Gly)	Gly allele raising D3 affinity & impulsivity	Weak and inconsistent

0.36

FTO β (kg/m²) strongest single anchor

13.7%

Chronotype SNP heritability vs 40–50% twin

18

Candidate-gene depression hypotheses, zero replicated

~30%

Non-zero floor where the genomic prior never fully decays

§ 6.0 From Attribution to Causation

The anchor delivers a genuine causal gain at Rung 1 of Pearl's ladder a calibrated, ranked hypothesis. Stronger claims require intervention or counterfactual evidence the framework is explicit about not having yet.

01

Rung 1 Attribution by Construction

Because Ĝ is exogenous, δ = P − Ĝ is non-genetic by construction. Whatever drives the deviation must lie in environment, state, or noise one confounder class eliminated structurally rather than by statistical adjustment. The output is a ranked, calibrated list of candidates, not a verdict.

02

Rung 2 Population & Individual Intervention

At the population level, the genotype serves as an instrumental variable for Mendelian randomization. At the individual level, an n-of-1 ABAB design introducing and withdrawing a candidate cause repeatedly while tracking δ against the stable genomic reference strengthens token causal claims.

03

Rung 3 The Honest Ceiling

The strongest single-reading output the framework will ever produce: a deviation magnitude, a ranked list of testable non-genetic candidates, and explicit confidence levels for the anchor, the exogeneity claim, and the ranking itself. Attribution, not proof and more useful than either a population "normal" or cold-start silence.

§ 7.0 Constraints for Honest Implementation

Calibrated restraint is the framework's defining discipline. The anchor is genuinely informative and genuinely weak, and every deployment constraint exists to keep the architecture from forgetting the second half of that sentence.

01

Weak & Evidence-Graded

Trait-specific uncertainty calibrated to evidence grade FTO and FADS provide strong priors; dopaminergic and serotonergic markers provide weak directional nudges that lower confidence in an alert rather than drive one.

02

Ancestry-Matched

Nearly every published effect size was estimated in European-ancestry cohorts. For autonomic, circadian, and dopaminergic domains, no South Asian GWAS summary statistics exist at all applying mismatched betas elsewhere systematically misplaces the anchor and compounds existing health disparities.

03

Dynamic & Epigenetically Aware

DNA sequence is fixed; its expression is not. FKBP5 shows the same genotype producing different effective set points depending on early-life history the framework must acknowledge it captures the DNA contribution only, with epigenetics shifting the true set point beneath it.

04

Attribution, Not Diagnosis

Outputs are ranked causal hypotheses, never diagnoses, and are explicitly unsuitable for clinical, employment, insurance, or forensic decisions. The MAOA "warrior gene" example is the cautionary tale: a weak G×E association with documented criminal-justice harms.

~/conclusion

$ query: what does the genetic anchor solve // the cold-start gap a day-zero reference before any behavioral data exist. // it doesn't predict a value. it's a weak, principled prior over one. $ query: what does this cost // a recoverable genetic variance that's small, polygenic, and ancestry-biased. // strongest for metabolic traits, weakest for anything behavioral. $ query: what is the actual result // the same HRV reading tells opposite causal stories for two different genomes. // population norms can't see that. an individualized anchor can.

THE GENOME
ISN'T A
VERDICT.
IT'S A
PRIOR.

IS IT YOU OR YOUR ENVIRONMENT?

IS IT YOU
OR YOUR
ENVIRONMENT?