Perfect batch confounding
When condition IS the batch
The stated claim
“GENE_1999 is up-regulated in condition A vs B — the effect looks strong and clean.”
benchmark case C1_batch_perfectDeciding evidence
Cramér's V between condition and batch = 1.00 — condition and batch are perfectly aliased.
Confound sentinel: batch/condition confounding. The evidence is a re-derivable association statistic, not a model opinion.
If every A sample was run in one batch and every B sample in another, the biological effect and the batch effect are mathematically indistinguishable. No amount of downstream modelling can separate them — the design itself is the problem.
What the engine does
The claim looks pristine
A large, consistent fold-change with a tight p-value. Nothing in the result table hints at a problem.
The sentinel measures the design
VERITAS computes the association between the condition label and the batch label. Cramér's V comes back at 1.00 — perfect aliasing.
The design-matrix batch model cannot rescue it
Adding batch to the design matrix (~condition + batch) yields a rank-deficient model: the effect is unidentifiable. There is no defensible analysis choice that separates the two.
Verdict: does not survive
A severe structural confound (V ≥ 0.9) forces RED. The certificate carries the exact statistic so anyone can re-derive it.
What the scientist does next
The scientist learns the experiment must be re-run with batches crossed against condition — the claim cannot be salvaged by analysis alone.
Verify your own finding
Run a bundled sample of this exact failure mode in the console, or open the validation atlas to see how it is graded.