FraudTrust Chaos Lab executable-model sample bundle Upload reference.csv and comparison.csv at /projects/new, together with logistic-model.json. All data are harmless synthetic fixture values. The manifest is data-only JSON: it declares a logistic intercept, numeric feature weights, exact feature ordering/typing, reject-on-missing policy and fraud-probability output semantics. The app mutates raw inputs, reruns this model, and derives exact signed linear contributions in logit space. The score column contains the expected model probability for every row. On upload, Chaos Lab independently recomputes each score and rejects the bundle if any supplied score differs by more than 1e-9. See data-dictionary.json for field definitions and checksums.txt for SHA-256 file digests.