Joint likelihood for correlated observations O-25, O-26, O-27 (shared basis: S-20). Step 8 writes the single ## Likelihood block here; the member edges point back via group.
Likelihood
# CG-32 (HC-12) — ONE joint estimate over members E-2 (O-25) + E-3 (O-26) + E-4 (O-27): all three are the same
# CDC surveillance analysis, shared basis S-20 (rule 1). Members in HC-12.hypotheses order [H-11 internal
# trans-ovarian route drives it, H-12 eggs a MAJOR vehicle (magnitude), H-58 residual negation]. NON-exhaustive:
# H-58 gets a real, argued likelihood (rule 3). Judge the whole pattern: 77% of vehicle-identified NE outbreaks
# traced to INTACT, uncracked, refrigerated Grade A eggs; SE egg-associated 44% vs 15% for other serotypes;
# bulk-raw-egg cultures 10% NE vs 0% elsewhere tracking illness geography. Anchored on H-11 = 1 (rule 7).
lik_se_H11 = 1.0 # anchor: the internal-route thesis entails eggs being the major vehicle (so it predicts all
# three magnitude signals) AND uniquely explains the distinctive detail that intact,
# refrigerated eggs still cause illness. A-7 (corrected): intact/refrigerated illness strongly
# disfavours external routes, making internal trans-ovarian the best-supported explanation
# (corrected from "excluded/forced" — external route disfavoured, not eliminated).
lik_se_H12 = 0.8 # ~0.8x the anchor: the pure magnitude claim predicts the 77% / 44-vs-15 / geography signals
# just as well, but is route-agnostic, so it is slightly more surprised by the
# intact-refrigerated detail (only H-11 puts full mass on the internal route). A-8 (approved):
# the serotype specificity (44 vs 15) ties the vehicle to SE — nailed equally by both facets.
lik_se_H58 = 0.15 # ~0.15x the anchor: the residual negation (external shell route controllable by handling, OR
# eggs not the principal vehicle) must reproduce THREE convergent, multi-stream signals at
# once — attribution, SE-serotype specificity, and direct egg-culture geography. External
# route is contradicted by intact/refrigerated illness; "not the vehicle" by the 77% and the
# cultures. Kept non-trivial (rule 3): an unlisted account — e.g. a shared upstream source —
# could partly reproduce the pattern.
t_se = 0.8 # cap = min trust_score over {O-25, O-26, O-27} = 0.83 (all via S-20). Docked lightly to 0.8
# for the vehicle-identification / ascertainment selection effect A-8 flags on the raw
# attribution figures (O-25/O-26); dock kept small because O-27 is direct microbiology immune
# to that bias and the signals span independent data streams (1973-84 national, 1985-87 NE,
# plant cultures).
evidence("HC-12", ["O-25", "O-26", "O-27"], [lik_se_H11, lik_se_H12, lik_se_H58], t=t_se)