These research notes are provided as-is and contain supplementary working research.

Analysis

The stopping rule and the eight self-checks, including the void-negative defect

The inventory is complete only when every instrument has been run for a father John Gurney and for each of the children Mary, Richard and John, variant-expanded, across 1615-1645 for England and Wales; positive controls pass on every instrument; every returned row is catalogued with its provenance and a household id; every household is scored with its shape hypothesis and discriminating evidence recorded; a coverage statement exists per county worked; and the residual is named as a list of parishes whose registers are known lost for 1620-1640, so the unsearchable remainder is a stated quantity rather than an unknown. Eight self-checks enforce it. Read the project's own record first, including the source registry. Run a positive control on every sweep - a query that does not return a record already known to exist and in scope is broken, and its result is not a finding. Match the place token to the target set rather than to the parish's name, because a keyword filter is a literal token match that fails closed. Check that the surname stem actually reaches the spelling in use, since a wildcard covers one stem and is not a variant sweep. Never combine a year bound with a place keyword in a single-dataset query: the combination fails closed, proven by 349 baptisms at one parish with no year bound and zero with a year bound in a set that demonstrably holds a baptism in that year under another surname - and any negative in the existing record resting on both parameters at once is void and must be re-run. Pre-register the expected order of magnitude. Record no negative from one provider alone, since no index is complete and the gaps do not coincide. Ask every question by child, by parent and by place. Weight the variant and never exclude a surname as a class. Weight against shape rather than volume. Enumerate shapes before collapsing a household to one. Stop and rebuild the method after two corrections in the same domain. And treat a finding without an instrument code and a query date as not a finding.

Confidence: High

Why this conclusion

Qualifications and conflicts

Related context

Sources

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

Relationship map

Technical details
  • Research item: G13-RI-000231
  • Graph revision: 381
  • Machine-readable JSON: /assets/g13-graph/findings/G13-RI-000231.json

This is one research item from the John Gurney research graph; the fuller narrative treatment is in the linked topic above and the research library.