Prediction readiness gate
Prediction is a separate scientific release, not a label upgrade.
No P1 or P2 prediction claim can be made until independent disease-relevant perturbation data, prospective registration, external test results, calibration, uncertainty, OOD and abstention analyses are frozen.
P0/P1/P2 ladder
What is unlocked today?
P0 · Prediction infrastructure onlyPASSregistry, schema, benchmark and abstention infrastructure exist
P1 · Research-use DMD perturbation-response predictionLOCKED_PENDING_EXTERNAL_VALIDATIONnone in this release
P2 · Clinical disease prediction or decision supportLOCKED_NOT_IN_SCOPEnone
Current denominators
The registry is intentionally empty
Governed hypotheses21
Observed HepG2 perturbations21
External myoblast context screens9
Independent muscle/DMD perturbation replications0
Registered predictions0
Measured outcomes0
Zero registered predictions and zero measured outcomes mean no qualifying prediction/outcome object exists; this is not a negative validation result.
Allowed now
P0 infrastructure claims
- static prediction-registration schema and empty registry are available
- same-HepG2 perturbation-response benchmark is auditable
- future prospective outcomes can be linked without overwriting predictions
- evidence gaps and abstention requirements are explicit
Locked
Claims not authorized
- validated DMD disease predictor
- therapeutic response predictor
- clinical decision-support platform
- treatment-prioritization engine
P1 unlock package
Minimum evidence before research-use DMD prediction
P1 requires independent disease-relevant perturbation validation before any research-use DMD prediction claim can be unlocked.
P1 required before unlock
- frozen DMD/dystrophin-deficient human myogenic or neuromuscular perturbation dataset
- train/validation/external-test partition declared before model fitting
- donor, isogenic-pair or perturbation unit of analysis declared
- negative, positive and rescue/control definitions predeclared
- primary functional endpoint and transcriptomic secondary endpoints frozen
- baseline models, calibration, uncertainty intervals and abstention policy benchmarked
- out-of-distribution detection and failure-mode taxonomy frozen
- prospective prediction cards timestamped before outcome access
- outcome board populated without relabelling failed or abstained predictions
P2 remains out of scope until
- clinical intended-use statement
- independent external clinical validation
- decision-impact or workflow evaluation protocol
- bias, safety, monitoring and human-oversight plan
- regulatory and ethics review appropriate to intended use
Reporting alignment. TRIPOD+AI-style model reporting fields are required before a P1 prediction claim DECIDE-AI/SPIRIT-AI/CONSORT-AI-style clinical evaluation fields are not triggered until a clinical workflow is proposed