A CXCL10-conditioned mechanistic external control for HTLV-1-associated myelopathy: blind reproduction of a registry control arm and design of the next disease-modifying trial (technical note and code, v1.0.0)

HTLV-1-associated myelopathy/tropical spastic paraparesis (HAM/TSP) progresses over years while trials last weeks to months. This record contains a technical note, an addendum and code for a minimal mechanistic model linking cerebrospinal fluid CXCL10 to the intensity of 1-grade motor worsening events, calibrated on published usual-care aggregates (no patient exposed to an investigational drug), and used as a model-based control. Version 1.1.0 adds three elements. (1) Prespecified, falsifiable predictions for patient-level data, frozen on 2026-11-01T03:31:04Z (SHA-256 hashes included) before any such data are requested. They give OMDS worsening at 1 and 4 years by CSF CXCL10 class, with and without a mogamulizumab-like CXCL10 reduction, and 10-m timed walk drift at 2 years by CXCL10 tertile. (2) Sensitivity analyses after a structured critical review: type I error with 5,000 null trials per design; 19 misspecified-model scenarios; observation-model variants (realized damage, CXCL10 screening error); recalibration of the 10-m timed walk link; a corrected time-to-worsening sample size that accounts for frailty attenuation of the hazard ratio. Prognostic covariate adjustment kept a type I error of 1.8-3.2% under every misspecification. Borrowing from the virtual arm and single-arm comparisons did not. (3) Corrections to the v1.0.0 note. The blinded check of the Brain 2023 control arm (38.4% predicted, 38.2% observed) tests the baseline progression rate, not the CXCL10 link, and the control cohort probably overlaps the calibration registry. Published aggregate data only; no patient-level data accessed.