Consistency: Deep research ↔ review YAML ↔ PN annotation consistent. EDEM3 = soluble ER-lumenal, calcium-dependent GH47 alpha-1,2-mannosidase (EC 3.2.1.113) catalyzing the downstream step Man8→Man7 (and toward Man5), with a unique PA domain; binds SEL1L (like EDEM1, unlike EDEM2); biallelic LoF causes EDEM3-CDG/CDG2V. Unlike EDEM1/EDEM2, EDEM3 has no negated mannosidase annotation — catalytic activity is undisputed (IBA/IEA/ISS/IMP + EC/RHEA all positive). No contradictions; this matches the prompt's note that EDEM3 is the demonstrably catalytic EDEM.
PN story / NEW pressure: subtype→GO:1904382 is already_in_goa_exact (TAS Reactome + review). EDEM3's role is fully captured (GO:0004571 across IBA/IEA/ISS/IMP/TAS; GO:1904380; GO:1904382; GO:0036503; GO:0097466). No NEW term needed; review proposes none. Already captured.
Mapping strategy: subtype→GO:1904382 exact and strongly supported — keep; EDEM3 is the prototypical "Mannose trimming" catalytic member. group→GO:0006487 protein N-linked glycosylation (new_to_goa) is broader/upstream (glycan biosynthesis) than EDEM3's degradative trimming and is a loose fit — borderline over-reach for a degradation-arm enzyme. The MF (GO:0004571) is correctly the clearest/strongest of the three EDEMs.
Evidence alignment: Strong overlap on PMID:25092655 (shared) and Reactome:R-HSA-6782685. Review adds rich EDEM3-specific literature (original characterization PMID:16431915 with E147Q catalytic proof; ERp46/TXNDC5 redox partner PMID:29784879; domain dissection PMID:33671632; HCC/HBV PMID:39838427) well beyond the terse PN row.
Verdict: Consistent and well-curated; the unambiguously catalytic EDEM. subtype→GO:1904382 sound and already captured; group→GO:0006487 is a loose/broad fit for the degradation arm.