H-17 - Eggs are a relatively unhealthy protein source; benefit comes from replacing eggs with plant or fish protein
The authors’ overall ranking: nuts and whole grains healthiest; eggs and processed meat unhealthiest; fish the single not-unhealthy animal source.
Link to original
H-18 - The egg-health effect is a property of the comparator not of eggs alone, so eggs good-or-bad is ill-posed
Motivated by the observation that the egg-removal effect ranges from null (vs processed meat) to strongly protective (vs nuts/whole grains) within a single dataset, and the authors’ caveat that an inverse substitution association does not by itself guarantee a healthy choice.
Link to original
H-23 - Egg's CVD impact is relative to the food it replaces (neutral vs healthy proteins, better than red-processed meat)
The substitution framing: eggs have no intrinsic fixed CVD effect; the relevant quantity is the difference from the displaced food.
Link to original
H-38 - The health effect of eggs depends on background diet, being least harmful when displacing refined carbohydrate
Advanced by the authors to explain why the US pooled cohort found harm while Kadoorie found protection and PURE found neutrality; also echoes the EAT-Lancet note that higher egg intake may benefit low-income populations with poor dietary quality.
Link to original
H-57 - Some other egg-health characterization
Residual catch-all so the comparator/well-posedness prior sums to 1; names an escape beyond intrinsic vs the contingency flavours.
Link to original
Carving
The sub-question is whether egg-health is an intrinsic fixed property of eggs or a property of the comparator and background diet - i.e. whether an unconditional verdict is even well-posed. H-17 asserts an intrinsic property: eggs are among the least-healthy protein sources and benefit accrues from replacing them. The contingency pole holds that there is no fixed effect - the verdict is a property of the displaced food and background diet - expressed generally (H-18, the unconditional verdict is ill-posed), as substitution-indexing (H-23, neutral versus fish/poultry/plant proteins, preferable to red/processed meat), and as background-diet-indexing (H-38, least harmful displacing refined carbohydrate, which also reconciles protective Asian and null/harmful Western findings). At most one best characterization holds. The contingency members are refinements at a shared grain, discriminated from H-17 by the large within-dataset substitution swings (egg-removal ranging from null vs processed meat to strongly protective vs nuts/whole grains). Residual H-57 covers a partly-intrinsic, partly-contingent mix. Kept distinct from HC-1 and HC-2 (this is a well-posedness/framing question, not a numeric sign or a geographic pattern), with the couplings logged in depends_on.
Prior
# HC-10 prior — is egg-health an intrinsic fixed property of eggs or a property of the comparator / background diet?
# Members in HC-10.hypotheses order: [H-17 intrinsic-unhealthy fixed rank; H-18 unconditional verdict ill-posed
# (fully comparator-contingent); H-23 substitution-indexed; H-38 background-diet-indexed; H-57 residual]. Non-exhaustive,
# residual last. Evidence split: all nine inbound E edges are EGG-SPECIFIC discriminating observations — the
# substitution rankings (E-88/89/90 in CG-30; E-87/91 in CG-29) and the region/diet-quality gradients (E-92/93/95 in
# CG-29, E-94) tell H-17 from the contingency framings and separate the contingency flavours — so NONE is marked
# used_for_prior; all are priced at step 8. no_observation_arguments.py --cluster HC-10 returns none. This prior is
# therefore a pure OUTSIDE VIEW on the framing question, resting on a methodological base rate, not on the egg data.
# H-17 intrinsic anchor — the unit weight. What must hold: eggs carry a comparator-STABLE low rank (a fixed property,
# benefit accruing from replacing them), not merely a rank that flips with the displaced food. Reference class: single
# foods whose health effect is claimed to be intrinsic rather than substitution-defined. This is the minority framing
# because the field's default (below) runs the other way, but eggs' distinctive cholesterol/choline load makes a
# stable-ish signal plausible, so it keeps real weight — comparable to each single contingency member.
w_intrinsic = 1.0 # H-17 anchor: a stable intrinsic egg rank is plausible but is the minority framing
# Contingency family (H-18 + H-23 + H-38) as odds vs the intrinsic pole. In nutritional epidemiology a single food's
# modelled effect is an isocaloric substitution quantity — the difference of two food coefficients — so "what does it
# replace / against what background" is near-definitionally load-bearing, and the field's default frame is contingency.
# Hence the contingency family outweighs the intrinsic pole several-fold a priori.
contingency_vs_intrinsic = 3.0 # would be right if the framing splits ~75% contingency / ~25% intrinsic across foods
w_contingency_family = contingency_vs_intrinsic * w_intrinsic
# Partition the contingency family across its three flavours (shares sum to 1 by construction). H-18 (general
# ill-posedness) is the broadest and most defensible, so the largest share; H-23 (substitution-indexing) is the
# standard analytic frame, nearly as broad; H-38 (background-diet-indexing) makes the most SPECIFIC directional claim
# (least harmful displacing refined carbohydrate, reconciling Asian-protective / Western-null), so the smallest share.
share_illposed = 0.40 # H-18: broadest contingency framing -> largest slice
share_substitution = 0.35 # H-23: substitution-indexing, the routine analytic frame
share_bgdiet = 0.25 # H-38: most specific directional claim -> smallest slice; 0.40+0.35+0.25 = 1.0
w_h18 = w_contingency_family * share_illposed
w_h23 = w_contingency_family * share_substitution
w_h38 = w_contingency_family * share_bgdiet
# H-57 residual — the truth is a characterization none of the four clean framings captures, most concretely a
# partly-intrinsic / partly-contingent hybrid (a stable egg component PLUS comparator/background modulation). Its own
# argued weight (rule 4), not a leftover: genuinely plausible for a messy real diet, but the four framings — H-18
# especially — already span most of the space, so it sits below each modelled member.
w_residual = 0.5 # ~0.5:1 vs the intrinsic anchor; ~11% of the mass once normalised
prior("HC-10", [w_intrinsic, w_h18, w_h23, w_h38, w_residual])