level_id	title	holdout	can_prove	cannot_prove	unlock_gate
E0	Pipeline execution	random cell or engineering smoke split	The adapter, data path and metric code execute.	Biological generalization.	Frozen task, complete receipts and no fatal schema error.
E1	Held-out perturbation	unseen perturbation in a matched context	Bounded perturbation interpolation or extrapolation in that context.	New-donor, new-cell-state or disease transfer.	Beat every legal permanent baseline across seeds on perturbation-specific metrics.
E2	Held-out donor	complete biological donor	Donor transfer within the frozen disease, state and assay.	New cell state or disease transfer.	Donor-disjoint split, donor-level uncertainty and no cell pseudo-replication.
E3	Held-out cell state	unseen cell state or cell type	Cross-state transfer for the frozen output contract.	Cross-disease or prospective validity.	State-disjoint evaluation and state-compatible baselines.
E4	Held-out disease or context	unseen disease or biological context	Retrospective cross-context transfer on a named dataset.	Future experimental performance.	Disease/context-disjoint truth with no pretraining exposure inheritance.
E5	External retrospective dataset	independent external dataset	External retrospective validity within the named dataset and endpoint.	Prospective or clinical validity.	Independent source, frozen preprocessing, external units and complete failure reporting.
E6	Prospective preregistered experiment	future independent donors or experiments	Prospective task-level evidence when the preregistered endpoint and baseline gate pass.	Clinical efficacy, treatment selection or patient-level validity.	Immutable Prediction before Outcome, independent experiment, calibration and all negative branches returned.