C.28.CM:4.6 - Challenge the consequence with available evidence

Compare the consequence with an observation, contrasting case, subject argument or feasible intervention that bears on it. First check that the observation concerns the modeled variable, cases and time. A contradiction can arise from the mechanism, measurement, implementation of a proposed action, or a changed operating condition.

Distinguish a contradicted implication, an untested implication and compatibility with the inspected material. A graph with no relevant testable implication cannot acquire support merely from an absence of contradiction. Conversely, failure to reject an implied independence does not establish the graph. Deterministic constraints, measurement error, missing cases and sampling uncertainty can all change the interpretation.

Revise only as far as the finding warrants. Preserve a valid observation when its causal interpretation fails. Restore a rival when its rejected premise changes. If an additional study could change the answer, C.28:4.8 helps specify the evidence question and C.11.DUA helps choose whether to obtain it. A useful model may end with an unresolved distinction.