Analysis

One row per record instance, and a household split is a hypothesis rather than a finding

Two disciplines govern how records become households. The catalogue takes one row per record instance, not per child: the same baptism indexed by four providers is four rows, because the disagreements between them are themselves evidence. Names are recorded exactly as indexed, since normalising to 'Gurney' destroys the variant evidence this work repeatedly needs, and no row enters without its instrument code and query date. Household assembly is where the July 2026 rounds failed hardest: four households were split on plausibility arguments and each split was then treated as a finding, when a split is a hypothesis about record shape and record shape is exactly what is uncertain. A household is therefore a set of catalogued rows plus an explicit shape hypothesis, never a bare assertion. Where a baptism run is too long for one father, the alternatives are enumerated rather than chosen between - one man with two wives; a father and son; two unrelated same-name men in one parish; an index conflating two parishes; a mis-transcribed father forename in some rows - and the evidence that would discriminate is recorded and the shape held open until it exists. No probability is built on an unresolved shape; where a row rests on a split, it is labelled conditional on that shape.

Confidence: High

Topic file location: Identity — method for a complete child inventory