layer_id	title	question	input	artifact	gate	local_route
TECH-01	Identity & ontology	Are genes, cells, diseases, perturbations and assays named consistently?	source identifiers + metadata	canonical IDs, ontology terms and ambiguity receipts	No silent alias, species or cell-type coercion	/resource/search
TECH-02	Dataset contract	Can a machine and a reviewer reconstruct what each matrix dimension means?	matrix + sample/cell + feature metadata	typed AnnData-shaped Dataset Card with provenance	X, obs, var, layers and uns roles are explicit	/resource/datasets/
TECH-03	QC & harmonization	Which technical effects were measured, corrected or left unresolved?	raw/count layers + donor, batch and assay metadata	QC receipt, excluded units and transformation lineage	Biological replicates remain the inferential unit	/resource/data-universe/
TECH-04	Task & split builder	What is predicted, and along which axis must it generalize?	context × perturbation × time × output contract	frozen estimand, split, leakage checks and holdout identity	Target, context and output are unseen exactly as declared	/resource/benchmarks/generalization/
TECH-05	Permanent baseline lane	Does the task require a complex model at all?	the identical training set, genes, split and metric code	zero-change, mean, linear and compatible neighbor results	Advanced models cannot advance without same-coverage controls	/resource/benchmarks/
TECH-06	Model adapter	Can a model family consume this contract and return the required object?	frozen task + versioned model/configuration	ModelRun with mean vector, population or uncertainty output	Architecture capability is not inferred from model name	/resource/models/
TECH-07	Evaluation & gate	Did the model recover perturbation-specific and biologically useful signal?	predictions + held-out truth + permanent baselines	six-level scorecard, uncertainty and abstention decision	One aggregate expression metric cannot unlock a claim	/resource/no-wet-lab-evaluation/
TECH-08	Registry & outcome return	Can the result be reproduced, challenged and updated by an experiment?	ModelRun, Study, Prediction and measured Outcome	immutable receipt and evidence-graph return	Null, toxic and failed outcomes are retained	/resource/registry