[pipeline] Commit reagent→role vocabulary table for rule-based role assignment

[rhowardstone]

Problem

Role assignment is currently LLM-only and 52% of reagents have null role. Many organoid reagents have fixed, well-known biological roles that are deterministic from name alone.

Vocabulary established from paper analysis

Chemical	Synonyms	Biological Role	Pathway
Y-27632	H1152, ROCK inhibitor	Prevents anoikis/dissociation	Rho-kinase
CHIR99021	GSK3β inhibitor, WNT activator	Self-renewal, WNT activation	Wnt/β-catenin
Dorsomorphin	BMP inhibitor, ALK2/3/6 inhibitor	Neural/mesoderm induction	BMP signaling
Noggin	BMP antagonist (conditioned)	Blocks osteogenesis	BMP signaling
SB431542	ALK4/5/7 inhibitor, TGF-β inhibitor	Promotes differentiation	TGF-β pathway
A83-01	ALK5/4 inhibitor	Modulates EMT	TGF-β pathway
FGF2/bFGF	Fibroblast growth factor 2	Proliferation, self-renewal	FGF signaling
FGF10	Fibroblast growth factor 10	Branching morphogenesis	FGF signaling
EGF	Epidermal growth factor	Progenitor expansion	EGFR signaling
BDNF	Brain-derived neurotrophic factor	Neuronal maturation (late)	TrkB pathway
NT3	Neurotrophin-3	Neuronal survival	TrkC pathway
Valproic acid	HDAC inhibitor, VPA	Chromatin remodeling	Histone acetylation
PGE2	Prostaglandin E2	Tissue regeneration	Prostaglandin pathway
Nicotinamide	Vitamin B3, NAD+ precursor	Proliferation, Wnt modulaton	NAD+ metabolism
N-acetyl-L-cysteine	NAC, antioxidant	Oxidative stress protection	Glutathione synthesis
Matrigel	BME, ECM	3D matrix/basement membrane	ECM signaling

Work (for Pipeline Engineer)

1. Commit pipeline/reagent_roles.json with the vocabulary above (expand from ChEMBL/PubChem)
2. In ground_reagents.py (or equivalent), apply exact + fuzzy name matching to assign role before LLM fallback
3. This reduces LLM calls and raises role coverage from 48% toward 90%+
4. Update reagents.jsonl export to reflect improved coverage

Gate: role populated for ≥85% of reagents in the next pipeline run.