Methods
Scoping a first record
Start with one contested entity and a bounded decision, then test whether its evidence can be defended.
Problem
A large corpus can make progress look substantial while leaving the central decision undefined. A first record is more useful when it exposes a real disagreement about one entity and a few consequential properties. The exercise tests whether a reader can trace each value, unknown and conflict to evidence and reproduce a past state. It does not require broad collection. Its success criterion is a defensible decision boundary, not the volume of documents gathered.
Steps
Choose a contested case
Select one entity for which sources disagree, use different identifiers or report values at different dates. Define the decision a reader must make and why disputed properties matter. Avoid a conveniently clean example that cannot test conflict handling. If the entity is fictional for a demonstration, label every fictional fact clearly as such.
Bound the record
List the few properties needed for the decision, permitted source classes, jurisdiction and time window. Define resolved, disputed and unknown outcomes. Set a stopping rule for search so the exercise does not expand into a census. A narrow boundary makes missing evidence visible and lets another team repeat the case without guessing what was in scope.
Assemble competing evidence
For each property, capture asserting passage, locator, reading date and retained copy where allowed. Keep candidate values, relevant absences and source dependencies separate. If documents disagree because they describe different dates or entities, state that explicitly. Include enough contrary evidence to test a decision rule, not merely enough supporting material to display a chosen answer.
Test the decision
Ask an independent reader to explain which value they would use, why, and what remains unknown. Replay the answer at an earlier cutoff if sources changed over time. Note where the reader needs a missing source, an access decision or a clearer rule. Use those findings to define the next bounded case; one record cannot establish corpus-wide performance.
Checks
- Decision, entity, properties and stopping rule are written down.
- Competing evidence and unknowns are visible.
- An independent reader can explain the choice and alternatives.
- The result is a case finding, not a population claim.
Pitfalls
- Starting with document volume rather than a decision.
- Choosing only a clean example.
- Generalizing from one case to every source class.
A worked example
A fictional buyer at a pharmaceutical distributor needs to decide whether to qualify a new supplier, Halden Laboratory Services GmbH. The decision boundary is written first: yes or no on qualification for one product category, within thirty days. Four properties matter: legal identity, manufacturing site location, most recent inspection outcome and licence status. Sources allowed: the national registers, the regulator's inspection database and the supplier's own certificates.
The assembled evidence shows the legal entity under two names, a site address that differs between the certificate and the inspection database, an inspection outcome from the previous year and no licence entry for one of the two countries. For each property the record shows the competing values, the passages, the reading dates and the declared unknowns. An independent reader examines it and says what they would decide and what they would still want to know. Their questions become the scope of the second record.
Variations and edge cases
If real cases cannot be shared, build a fictional one with the same structure and label it so. A good fictional case has conflicting sources, a late correction, a restricted document and an unknown. It should be small enough to review in an hour. If the case is real, choose one where the decision is still open, not one already settled, since a settled case cannot test whether the record changes the decision.
Scope creep is the usual failure. A request to also add sibling suppliers, more properties and more jurisdictions arrives early. Resist adding them to the first record. Write them down for the next one. A first record that is small enough to finish is also small enough to criticize, and criticism is what it is for.
What the exercise tests
The aim is to learn whether the approach produces something a reader can trust and use, and which of the method's pieces matter for this reader: citations, dated readings, declared unknowns, conflict reports, replay. Measure what can be measured without pretending to a sample: time for the independent reader to reach a decision, number of questions they needed answered, and number of claims they could trace to a passage without help. Report these as observations from one case.
How the method is checked
Compare the written scope with what was built. Anything in the record outside the stated properties and sources is scope creep. Anything required by the decision and missing from the record is a gap. Check that the stopping rule was applied. Check that the reader's questions are written down and traceable to the next steps.
See also
- Declaring unknowns
- Reporting source conflicts
- Reproducing a past state
- Data catalog vs system of record
- System of record
- Provenance
Further reading in the articles
Found an error on this page? See corrections for what counts as one, how to report it and the dated log.
Enquiry
Test this method on one entity.
Name an entity where "Scoping a first record" would be checked in practice, with its sources and its readers. We reply with what a first record would cover and what it would need from you.
Or write to [email protected] in your own words.