torchcell.verification

Record-level verification of built datasets at five levels, L0 to L4. levels holds the reusable checks and common the rules every dataset family shares; each family module (fitness, expression, morphology, metabolite, protein and others) applies them to one phenotype type. report defines the pydantic verification report, sourced binds a single extracted value to its source quote and sha256, and runners runs the verifiers over built datasets.

Classes

CarrierGapCensus	The gap census, extended to every carrier and to undeclared None values.
SharedRecordRules	Accumulate the family-agnostic rules over a dataset's records.
DerivationMethod	How a non-sourced statistic's value was fixed (ordered by preference).
Level	The five universal verification levels (ordered: a gate, not a set).
LevelResult	The outcome of a single L-level check.
Provenance	Where a dataset's records came from, recorded alongside verification.
StatDerivation	How a derived (non-per-record-sourced) statistic was fixed, as data.
VerificationReport	Provenance + the L-level results for one dataset.
SourcedValue	A hardcoded value plus the provenance that justifies it.
ProvenanceGap	The complement of SourcedValue: a documented, typed ABSENCE.
ProvenanceGapReason	WHERE in the provenance chain a value died -- an honest, typed absence.
ProvenanceGapCensus	Aggregate view of the ProvenanceGaps across a built dataset -- the worklist.

Functions

shared_rule_results	Run every shared rule over a materialized record sequence.
sha256_file	Return the hex sha256 of a file, read in chunks (large artifacts).
audit_sourced_value	Verify a SourcedValue against its source artifact.
library_available	True if the MinerU library root exists (audits skip when it doesn't).
provenance_gap_census	Tally the provenance_gaps carried by a stream of phenotype records.
provenance_gap_level_result	Wrap a census into the L1 provenance_gaps result -- ALWAYS passes.
l1_provenance_gaps	L1 informational census of documented provenance gaps over a phenotype stream.
l0_structural	L0: every raw record validates against its schema model.
l1_completeness	L1: observed keys cover the expected oracle set (nothing silently dropped).
l1_count	L1 variant: an exact record-count oracle (e.g. 4718 mutants + 122 WT).
l2_cross_method	L2: two independently-derived aligned series agree within tol.
l2_value_fidelity	L2: numeric values are finite / in range / non-NaN as required.
l3_convention	L3: a named units/convention assertion (e.g. expression = log2(sample/ref)).
l4_cross_source	L4: overlapping entities agree across two sources.
measured_gene_universe	Union of all measured genes across records (the L4 cross-source key).
verify_expression_dataset	Run the L0-L4 record-level gate for one microarray expression dataset.
perturbed_gene_set	Union of systematic gene names deleted across records (the L4 overlap key).
verify_morphology_dataset	Run the L0-L3 record-level gate for the Ohya2005 CalMorph dataset.
verify_visual_score_dataset	Run the L0-L3 record-level gate for a visual-score dataset.
visual_score_gene_set	The host loci a dataset's records name: the L4 gene-universe overlap key.
metabolite_gene_set	The host loci a dataset's records name: the L4 gene-universe overlap key.
verify_metabolite_dataset	Run the L0-L3 record-level gate for a metabolite dataset.
fitness_gene_set	Union of screened deleted gene names -- the L4 gene-containment key.
verify_fitness_dataset	Run the L0-L4 record-level gate for a single-mutant fitness dataset.