Joint likelihood for correlated observations O-75, O-76, O-77, O-78, O-80 (shared basis: D-22). Step 8 writes the single ## Likelihood block here; the member edges point back via group.
Likelihood
# CG-21 (HC-6) — ONE joint estimate over O-75 + O-76 + O-77 + O-78 + O-80, all from Ginsberg 1994 (S-42,
# shared basis D-22), rule 1. Pattern whole: an 0/1/2/4-egg crossover in healthy young men gave a linear
# fasting rise of TC (+1.47/100mg) and LDL (+1.38/100mg, P<0.001) with NO HDL/TG dose response, apoB up
# ~10%, and responsiveness wide, normally distributed, mostly positive. Members in HC-6.hypotheses order
# [H-20, H-34, H-40, H-53]. Anchored on H-34 (rule 7): this IS its definition — a linear ~1.4-1.5/100mg
# absolute rise borne by LDL/apoB, not HDL.
lik_ginsberg_H34 = 1.0 # anchor: the linear dose-response + apoB rise + HDL-flat is H-34's exact
# fingerprint. A-27 (approved): the parallel ~10% apoB rise means added particles
# (atherogenic particle number), the specific H-34 claim
lik_ginsberg_H40 = 0.12 # 0.12×: strongly against — LDL rose (not preserved), HDL flat (not up), and the
# mostly-positive spread INCLUDING high responders contradicts H-40's "neutral
# even in hyper-absorbers" (E-25). Not near-zero: Ginsberg's non-CRD background
# sits outside H-40's carb-restricted scope, its only defence
lik_ginsberg_H20 = 0.3 # 0.3×: H-20 is on the same "raises atherogenic lipids" side but claims only a
# comparator-relative effect with LDL falling absolutely — a clean absolute linear
# rise is against that specific "no absolute rise" claim, though H-20 doesn't
# forbid a dose-response in a different (healthy, high-dose) population
lik_ginsberg_H53 = 0.35 # 0.35×: unlisted effect — the apoB rise specifically argues against an H-53
# particle-quality-ONLY effect that would leave apoB flat (E-24), so its most
# prominent flavour is disfavoured; middling overall (rule 3)
t_ginsberg = 0.8 # cap = min trust over the five obs = S-42 = 0.82; docked a hair for n=20 making
# O-80's distributional shape and the exact slopes less certain. Core effects are
# P<0.001 and robust, so near cap
evidence("HC-6", ["O-75", "O-76", "O-77", "O-78", "O-80"], [lik_ginsberg_H20, lik_ginsberg_H34, lik_ginsberg_H40, lik_ginsberg_H53], t=t_ginsberg)