D. Sengupta, S. Panda
Biomedical evidence can become mechanistically informative before it becomes sufficiently commensurate to calibrate one system model. We developed evidence-constrained mechanistic synthesis (ECMS), a pre-calibration framework that converts only the mathematical information supported by each finding into constraints on a possibility space of executable mechanistic worlds. We reconstructed historical ECMS systems for alopecia areata (AA) and chronic spontaneous urticaria (CSU) using knowledge available by 31 December 2023, froze molecule/regimen inputs, observation semantics and finite ensembles before outcome reveal, and evaluated later studies at prespecified information-property resolution rather than as calibrated clinical predictions. Among 41 registered atomic properties, ECMS committed to 34 executable propositions: 29 were empirically concordant, one was clearly contradicted, and four could not be fully assessed because intermediate reporting or intervention isolation was insufficient; seven additional properties were prespecified abstentions and were not counted as empirical successes. An evidence-lineage audit further separated nine direct historical analogues from 16 near transfers, four mechanistic transfers and four cross-stream syntheses. Baricitinib withdrawal in AA was a cross-stream synthesis: no pre-cutoff baricitinib-withdrawal result was used, yet the frozen system reproduced early persistence followed by progressive loss. Selective MRGPRX2 antagonism in CSU transferred from pre-cutoff route-separation evidence, whereas patient-level clinical efficacy was withheld. A later baricitinib trajectory contradicted the frozen week-36-to-52 monotonicity property, preserving falsifiability. These results show that heterogeneous pre-calibration evidence can be made executable and externally challenged without being converted into unsupported coefficients,posterior biological probabilities or patient-response models.