{ "evaluation_schema": "nmd-vcell-no-new-wet-lab-evaluation/1.0", "object_id": "NWLE:v1.0.0-database-resource:EA-20260729-20", "resource_release": "v1.0.0-database-resource", "evidence_freeze": "2026-07-25", "interface_build": "EA-20260729-20", "lifecycle": { "state": "RELEASED_EVALUATION_CONTRACT_PARTIALLY_EXECUTED", "new_wet_lab_required": false, "module_count": 11, "executed_or_available_modules": 4, "partial_modules": 3, "registered_not_executed_modules": 4, "new_dmd_functional_outcomes": 0 }, "purpose": "Use released data, public data and new computation to stress-test, falsify, downgrade and prioritize hypotheses before any new wet-lab study.", "distinction": "No new wet lab does not mean no new analysis or no new data ingestion. Every added public dataset and every new run must retain source, freeze, split, baseline, metric and uncertainty provenance.", "permitted_decisions": [ "KEEP_FOR_COMPUTATIONAL_TRIAGE", "DOWNGRADE", "ABSTAIN", "ESCALATE_TO_EXPERIMENT" ], "claim_ladder": [ { "level": "C0", "label": "Release integrity", "unlocks": "Reproducible access to the stated objects and routes.", "current_state": "AVAILABLE" }, { "level": "C1", "label": "Same-context computational support", "unlocks": "Bounded performance statements inside the frozen HepG2 evaluation context.", "current_state": "PARTIAL_STRICT_GATE_NOT_PASSED" }, { "level": "C2", "label": "External compatibility", "unlocks": "Compatibility or falsification statements on explicitly named public datasets.", "current_state": "PARTIAL_NO_DIRECTIONAL_TRANSFER_SUPPORT" }, { "level": "C3", "label": "DMD perturbation prediction", "unlocks": "Disease-background perturbation-response prediction.", "current_state": "LOCKED_DMD_PERTURBATION_TRUTH_MISSING" }, { "level": "C4", "label": "Causal function, safety or treatment utility", "unlocks": "Replicated DMD functional, safety or therapeutic claims.", "current_state": "LOCKED_REQUIRES_EXPERIMENTAL_OUTCOME" } ], "modules": [ { "module_id": "NWLE-00", "title": "Release and provenance integrity", "question": "Can every displayed claim, derived object and download be traced to a versioned source and checksum?", "independent_unit": "release object or rendered route", "baselines": [ "schema conformance", "checksum match", "route manifest" ], "primary_outputs": [ "release identity consistency", "machine-readable provenance", "route and download integrity" ], "current_state": "RELEASED_EXECUTED", "claim_impact": "Supports reproducibility of the released resource; it does not support a biological claim.", "evidence": [ "/resource/api/v1.1/release_manifest.json", "/resource/api/v1.1/overlay_archive_identity.json", "/resource/qa-status" ] }, { "module_id": "NWLE-01", "title": "Same-context predictive audit", "question": "Does the fitted model improve held-out same-HepG2 perturbation prediction over zero-change and training-mean baselines?", "independent_unit": "held-out perturbation target within repeated balanced folds", "baselines": [ "zero_change", "training_response_mean", "ridge_residual" ], "primary_outputs": [ "paired RMSE differences", "raw and perturbation-specific residual direction metrics", "multiplicity-aware strict gate" ], "current_state": "EXECUTED_BOUNDED_STRICT_GATE_0_OF_16", "claim_impact": "Supports a small same-context RMSE improvement only; strict direct-head support is not established.", "evidence": [ "/resource/api/v1.1/benchmark_summary.json", "/resource/api/v1.1/strict_outcomes.json", "/resource/benchmarks.html" ] }, { "module_id": "NWLE-02", "title": "External aggregate transfer diagnostic", "question": "Does a HepG2-trained response model transfer directional signal to an external perturbation panel?", "independent_unit": "external perturbation target", "baselines": [ "zero_change", "training_response_mean", "safe_ridge_transfer" ], "primary_outputs": [ "MAE and RMSE on aggregate deltas", "delta correlation", "L1 perturbation-discrimination score", "target-bootstrap intervals" ], "current_state": "EXECUTED_NO_DIRECTIONAL_TRANSFER_SUPPORT", "claim_impact": "Falsifies a broad directional-transfer claim on the current 55-target panel; it is not DMD or muscle truth.", "evidence": [ "/resource/api/v1.1/vcc_metric_diagnostic.json", "/resource/benchmarks/vcc-readiness/" ] }, { "module_id": "NWLE-03", "title": "Public myogenic and DMD context triangulation", "question": "Are candidate records compatible with public myogenic fusion, differentiation and DMD-correction reference layers?", "independent_unit": "public study, target or reference trajectory", "baselines": [ "study-specific negative controls", "candidate-library coverage", "unperturbed trajectory reference" ], "primary_outputs": [ "candidate coverage", "context compatibility", "direction conflict", "assay and disease-context gaps" ], "current_state": "EXECUTED_REFERENCE_ONLY", "claim_impact": "Can downgrade or contextualize hypotheses; cannot create independent DMD candidate-perturbation validation.", "evidence": [ "/resource/api/v1.1/external_perturbation_evidence.json", "/resource/applicability" ] }, { "module_id": "NWLE-04", "title": "Source-balanced DMD prior multiverse", "question": "Does candidate direction persist across source-balanced rules, compartments and leave-one-source-out analyses?", "independent_unit": "source-aware gene-context estimate", "baselines": [ "legacy q-weighted prior", "unweighted sign rule" ], "primary_outputs": [ "source-balanced direction", "near-zero class", "R1–R5 robustness", "leave-one-source flip count" ], "current_state": "REGISTERED_NOT_EXECUTED", "claim_impact": "Will improve disease-prior robustness, but still will not constitute perturbation efficacy.", "evidence": [ "/resource/dmd_prior" ] }, { "module_id": "NWLE-05", "title": "Leakage-safe G2 backtest", "question": "Does prediction persist when an entire gene family or pathway is excluded from training?", "independent_unit": "held-out gene family or pathway", "baselines": [ "zero_change", "training_response_mean", "nearest_neighbour_transfer", "pathway_module_mean" ], "primary_outputs": [ "family-level performance", "worst-decile performance", "baseline win rate", "leakage audit" ], "current_state": "REGISTERED_NOT_EXECUTED", "claim_impact": "Required before an unseen-family generalization claim.", "evidence": [ "/resource/api/v1.1/generalization_ladder.json", "/resource/benchmarks/generalization/" ] }, { "module_id": "NWLE-06", "title": "In-silico perturbation-mode boundary", "question": "Which single and paired perturbation previews are observed, predicted, additive nulls or unsupported?", "independent_unit": "perturbation × context × modality", "baselines": [ "observed CRISPRi", "additive no-interaction null" ], "primary_outputs": [ "simulation fidelity label", "supporting dataset", "interaction-identifiability flag", "locked unsupported modes" ], "current_state": "PARTIAL_BOUNDED_SIMULATION_ONLY", "claim_impact": "Observed same-HepG2 CRISPRi and additive null previews are usable; KO, overexpression, fate and synergy claims remain locked.", "evidence": [ "/resource/perturbation-lab", "/resource/api/v1.1/perturbation_simulation_contract.json" ] }, { "module_id": "NWLE-07", "title": "Uncertainty, OOD and abstention audit", "question": "Does uncertainty track risk, and can the system abstain on unsupported contexts?", "independent_unit": "target or context-level prediction", "baselines": [ "unselective prediction", "constant-width interval" ], "primary_outputs": [ "bootstrap interval coverage", "risk-coverage curve", "OOD score", "abstention utility" ], "current_state": "PARTIAL_BOOTSTRAP_AVAILABLE_OOD_ABSTENTION_NOT_TRAINED", "claim_impact": "Sampling intervals are available for selected executed analyses; trained OOD-aware abstention is not yet supported.", "evidence": [ "/resource/api/v1.1/benchmark_summary.json", "/resource/api/v1.1/generalization_ladder.json" ] }, { "module_id": "NWLE-08", "title": "Retrospective selection-enrichment backtest", "question": "Would a historically frozen NMD-VCell selection rule enrich public positives over random and simple baselines?", "independent_unit": "outcome-blinded candidate selection set", "baselines": [ "random selection", "simple ridge selection", "expert heuristic" ], "primary_outputs": [ "top-k hit rate", "enrichment over random", "calibration", "outcome-leakage audit" ], "current_state": "REGISTERED_NOT_EXECUTED_OUTCOME_LEAKAGE_GUARD", "claim_impact": "Only valid when a pre-outcome model snapshot and eligible outcome universe can be proven.", "evidence": [ "/resource/api/v1.1/prediction_registry.json", "/resource/registry" ] }, { "module_id": "NWLE-09", "title": "Independent computational reproduction", "question": "Can an external analyst reproduce the released scorecards from the portable objects without private state?", "independent_unit": "independent analyst execution", "baselines": [ "published checksum", "golden expected-output fixture" ], "primary_outputs": [ "environment report", "checksum agreement", "metric agreement", "documented discrepancy" ], "current_state": "REGISTERED_EXTERNAL_REPRODUCTION_NOT_EXECUTED", "claim_impact": "Independent reproduction can strengthen computational reliability, but not biological validity.", "evidence": [ "/resource/downloads", "/resource/api/v1.1/portable_release_bundle.json" ] }, { "module_id": "NWLE-10", "title": "UniPert cross-domain representation feasibility", "question": "Do frozen UniPert gene and compound representations add leakage-safe value over simple baselines for held-out genes, families, pathways or chemical scaffolds?", "independent_unit": "candidate gene, held-out target family or compound scaffold", "baselines": [ "ridge_native", "ESM2_only", "ECFP4_only", "nearest_neighbour_transfer", "random_embedding" ], "primary_outputs": [ "identifier and encoder coverage", "known-relation retrieval", "G2 incremental performance", "public cross-domain learning curve", "training-exposure and licence audit" ], "current_state": "PARTIAL_SOURCE_AND_IDENTIFIER_AUDIT_EXECUTED_MODEL_EVALUATION_PENDING", "claim_impact": "Supports operational feasibility only; no embedding benefit, cross-domain transfer or DMD therapeutic value has been established.", "evidence": [ "/resource/unipert-feasibility/", "/resource/api/v1.1/unipert_feasibility.json", "/resource/downloads/unipert_candidate_identifier_preflight.tsv" ] } ], "required_scorecard_fields": [ "evaluation_id", "module_id", "freeze_id", "dataset_ids", "eligible_universe", "independent_unit", "split_id", "model_id", "baseline_ids", "metric_id", "estimate", "interval", "multiplicity_rule", "result_state", "claim_impact", "artifact_uri", "executed_at" ], "prohibited_inferences": [ "Do not represent the negative of knockdown as overexpression truth.", "Do not represent an additive double-perturbation null as synergy prediction.", "Do not represent transcriptomic reversal as DMD functional rescue.", "Do not represent a public reference layer as independent candidate perturbation validation.", "Do not treat cells, images or spots within one biological sample as independent biological replicates.", "Do not promote computational compatibility to clinical, safety or treatment utility.", "Do not treat a protein or compound identity embedding as perturbation direction, dose, time, mechanism or DMD response." ], "exit_condition": "The pack may prioritize, reject or defer hypotheses. Promotion to DMD-validated evidence remains blocked until a registered experiment returns target engagement, functional, toxicity and replication outcomes.", "boundary": "This package adds no wet-lab experiment and no DMD functional outcome. Its highest valid use is auditable computational triage, external compatibility testing, falsification and experiment prioritization." }