What the analysis says
The cluster asks whether elevated plasma TMAO is itself causally atherogenic in humans or a non-causal marker, read as three exclusive rungs (highest-that-holds, so exhaustive_by_construction, no residual): H-15 - Gut flora convert dietary phosphatidylcholine-choline to TMAO, which causally promotes atherosclerosis independent of lipids (real gut-flora mechanism, human translation open), H-16 - Elevated plasma TMAO is a clinically meaningful (causal) cardiovascular risk factor in humans (established human causality), and the step-4-constructed null H-45 - TMAO is a non-causal marker not itself atherogenic in humans. The prior is a pure outside view treating TMAO as an unadjudicated CVD-associated metabolite, so the marker held plurality: [0.327, 0.218, 0.455] (H-15/H-16/H-45), leaning on an overturned-biomarker reference class (homocysteine, CRP, HDL, vitamin E, beta-carotene). Posterior [0.283, 0.396, 0.321]: human-causal H-16 rose from 0.218 to plurality 0.396, the marker fell 0.455 to 0.321, and the three-way split stayed open (nothing clears 0.4). Three movers, all Hazen-corpus, all at trust 0.72: CG-24 - HC-8 joint over O-32+O-33+O-35+O-36 (Wang 2011 mouse chain: antibiotic/germ-free necessity, TMAO-feeding sufficiency, FMO3 genetics) priced the marker at 0.5x, halving its odds against the mechanism poles; CG-25 - HC-8 joint over O-37+O-38 (GeneBank angiography associations) lifted H-16/H-45 over H-15; CG-26 - HC-8 joint over O-63+O-66 (free choline raising TMAO 6-fold plus a within-cohort platelet-reactivity coupling) uniquely favoured H-16 (1.0 vs 0.6/0.6) and did most of the work lifting it to plurality.
What the model may not capture
Is TMAO’s causal status a false trichotomy? The three members sit on one scalar ladder, but the true answer may be off it: TMAO could be causal only in a bounded regime (renal impairment; via platelets, not foam cells) while a marker elsewhere — a both/and the exclusive rungs cannot hold, forced onto whichever single rung “holds highest” and thereby inflating H-16. An unlisted “bounded-causal” answer would be more consequential than any listed one, because it changes for-whom, not just whether. The model also prices mouse-to-human translation as a member-discrimination ratio (H-45 at 0.5x/0.9x), so the strongest evidence — the clean causal mouse chain — barely moves the human question by construction, leaving H-16’s plurality resting almost entirely on one intermediate surrogate (platelet coupling, CG-26), not a hard outcome. COI cuts hard: all three movers are Hazen-lab, and Hazen holds TMAO-diagnostic patent/royalty interests (Cleveland HeartLab/Quest) — a financial stake in TMAO being established as clinically meaningful, i.e. motivated toward exactly the H-16 rise shown. The discriminating counter-evidence for H-45 (Mendelian randomization, renal-adjusted associations) is not in the curated set, and no pooled source had an incentive to press the marker reading.
What would help
- Mendelian-randomization of genetically-predicted TMAO vs CVD — the decisive H-16-vs-H-45 test, and the tool that overturned each of the prior’s analogues — exists, unread.
- Human TMAO-CVD associations adjusted for renal function; CG-25’s were not (the model flags this) — exists, unread.
- Non-Hazen replication of the human associations and platelet coupling — exists, unread.
- Whether TMAO is causal in a bounded subgroup while a marker generally (tests the trichotomy itself) — does not exist.
Confusions and contradictions
The mouse chain is about as clean a causal demonstration as biology offers (necessity plus sufficiency plus genetics), yet the human evidence is entirely observational or surrogate, and the two cannot be reconciled into a verdict — the model honestly encodes this as translation uncertainty rather than papering it, which is why no member clears 0.4. The irreducible knot: the single lab producing the causal mouse mechanism, the human associations, and the platelet coupling holds a diagnostic COI motivating the causal reading — yet that same lab’s Wilcox RCT (HC-9) argues eggs do not raise TMAO. The COI therefore does not point one way, and the analysis cannot tell whether H-16’s plurality is signal or motivated source-selection. Shipped unfixed, deliberately.