pCite

METABOLOMICS KNOWLEDGE GRAPH


Papers: 1,287
Claims: 8,761 (unique assertions, deduplicated)
Physical claims: 5,470 (MetaboLights deposit verified)

EXPERIMENT RESULTS


Hypothesis: pCite surfaces physically-validated claims better than traditional citation count.

Mann-Whitney Up = 0.00e+00   ✓ significant
Precision@50pCite 0.94 vs Traditional 0.50 (1.9× lift)
NDCG@50pCite 0.94 vs Traditional 0.60

Hypothesis holds.

VALIDATION CLASS DISTRIBUTION


Physical5,470   62.4%
Replicated25   0.3%
DB Referenced2,661   30.4%
Text Derived605   6.9%

The distribution is the argument. 7% of extracted claims have no physical anchor. Under traditional citation metrics, these claims are indistinguishable from the 62.4% that do.

CODE & DATA


GitHubgithub.com/VibeCodingScientist/pCite
DataCC0 — download claims.jsonl