Evidence
Understand the model and its boundaries.
The studio connects AI models with explicit data contracts, reproducible measurements and human review. A scientific result does not remove uncertainty about the meaning or permitted use of data.
The model used by the interactive clinical demo
Specialist treatments are separate experiments
Age quantiles and conditional-Beta PD-L1 were evaluated on artificial fixtures. Do not infer a joint Beta distribution or activation in the public checkpoint from the presence of these experiments.
Human-governed interpretation
Data meaning, permitted use and subgroup targets.
Model outputs and pre/post-processing behaviour.
Approve interpretation and any subsequent deployment.
Temporal meaning cannot be assumed
The frozen stage–metastasis audit remains unresolved under the current schema. A technical rule trigger is not a clinical contradiction; timestamps or authoritative definitions are needed before revising the rule.
Safety commitments
- Artificial clinical examples and licensed public imaging evidence; no arbitrary patient upload in this interface.
- Official data and sensitive derivatives require the authorised Secure Processing Environment (SPE), approved semantics and separate permissions.
- Missing, unknown and not-assessed values are distinct; unresolved mappings stop training.
- Exact-match, DCR and NNDR diagnostics reduce selected risks but do not prove anonymity or differential privacy.
- Observed subgroup harm cannot be compensated by a favourable overall average.
- No diagnostic, treatment, clinical-validity or causal-benefit claim follows from these demonstrations.