A pre-specified mineralization-inhibitor score separates systemic sclerosis fibroblasts within one study and does not replicate in an independent series

A six-gene mineralization-inhibitor score (ENPP1, ANKH, MGP, GGCX, VKORC1, NT5E) in CD90+ dermal fibroblasts reads lower in systemic sclerosis than in controls. The gene set, direction, detection floors and significance threshold were fixed before any expression value was read. In a pre-registered held-out set of 50 donors (25 and 25), the observed difference is -0.2939 at one-sided p = 0.0204. That replication is across platforms within one study (GSE195452), not across studies, and no ambient RNA correction was applied. The score did not replicate in an independent series. In GSE292979, the same frozen genes, direction, floors and threshold were applied under a pre-registration, and a blinded second implementation reproduced every donor score within 1e-6 before the result was unmasked. All four pre-registered contrasts were null (primary -0.0412, one-sided p 0.4955; minimum attainable p 0.0045). The paper offers no explanation for the null and does not claim the effect is absent. The score is a pre-specified composite, not a pathway readout: a leave-one-out shows the signal concentrated in VKORC1, GGCX and NT5E, and the paper makes no mechanistic claim. The osteogenic half of the original hypothesis did not clear its pre-registered detection floor in any dataset examined. This is not a study of calcinosis lesions, not evidence of specificity, and not a mechanism. It is a pre-registered result that held within one study and did not replicate in the one independent series tested, reported with both outcomes.