track_id	title	outcome	lessons
KNOW-01	Data truth	Read a matrix as a biological measurement contract, not an anonymous tensor.	What is the inferential unit?: cells are observations nested in donors or cultures [failure: pseudo-replication] /resource/datasets/ | What does each AnnData slot mean?: X + obs + var + layers + uns form one typed object [failure: opaque transformed inputs] /resource/data-universe/ | Which context is missing?: storage presence is not task qualification [failure: using DMD context as DMD perturbation outcome] /resource/prediction_readiness
KNOW-02	Model mechanisms	Choose an architecture from its assumptions, input and output—not from its reputation.	Mean vector or cell population?: the output object determines the model family [failure: calling an aggregate vector a virtual-cell population] /resource/models/ | Where does context enter?: covariate, graph, prompt and transport encode different assumptions [failure: architecture-name inference] /resource/technology-methods/#model-families | What was actually executed?: paper capability and local ModelRun are separate ledgers [failure: inheriting external performance] /resource/model-run/mrun-ridge-safe-2-3-g0-repeated-fold/
KNOW-03	Evaluation without shortcuts	Detect trivial mean predictors, leakage, mode collapse and overconfident extrapolation.	Why simple baselines are permanent: systematic variation can make trivial predictors look strong [failure: complexity bias] /resource/benchmarks/ | Why one metric is insufficient: mean, delta, distribution and function answer different questions [failure: metric substitution] /resource/no-wet-lab-evaluation/ | How OOD axes change the claim: unseen perturbation, donor and disease are different tasks [failure: generalization collapse] /resource/benchmarks/generalization/
KNOW-04	From prediction to experiment	Convert uncertainty or abstention into a study whose outcome can update the evidence graph.	Freeze the estimand: context × perturbation × time × endpoint is the scientific unit [failure: post-hoc endpoint switching] /resource/planner | Return every outcome: positive, null, toxic and failed results all reduce uncertainty [failure: success-only evidence] /resource/registry | Know the final claim ceiling: functional validation is not clinical validation [failure: therapeutic overclaim] /resource/claim_boundary