# Check whether research sources are independent

New Eruditions · Reusable worksheet

## Start with the claim

Write the specific assertion being checked. Record its measure, value, unit, period, place, population, exclusions, precision and whether it is a count, estimate, forecast or attributed view. Split a volume, share and causal explanation into separate claim IDs.

## Record each document

Use documents-template.csv to preserve the public URL, publisher, author, displayed publication/update dates, actual access date, version identity and access status. A current page is not proof of its first-day wording. If only an excerpt is available, label the record accordingly.

## Link evidence to the claim

For every claim × document combination, use claim-support-template.csv. Find the empirical inputs supporting that passage. Record actual dataset/report links and credits. Describe data origin, exact dataset version, calculations and reporting origin separately. Put the supported relationship in dependency-edges-template.csv.

Assign one or more claim-level classes:

- DER: explicit derivative or republication of an identified report
- COM: documented shared observations or dataset
- IND: separately evidenced collection, relevant to the focal claim
- PAR: partly shared and partly distinct inputs; name each part
- CTX: additional context for another claim, not another focal measurement
- UNK: origin or dependency unresolved

Keep relationship certainty separate: CONFIRMED, INFERRED, ACCESS-QUALIFIED or UNRESOLVED. An explicit ultimate-producer credit does not prove the writer's actual acquisition route.

## Group and count carefully

Define the grouping rule before the count. Group reports only where documented origins support shared observations. Matching numbers, timing, domain names or ownership are clues, not enough alone. Unknown accounts are neither automatically independent nor automatically merged.

Count selected downstream appearances, primary-origin records, identified observation families and unresolved assignments separately. Keep interview claims and analytical checks separate. Never add these unlike quantities into an effective-source score.

## Test the number

Reproduce valid transformations from identified input records. Retain category exclusions and stated denominators. Check rounding and percentage-point differences. A fresh calculation of shared data checks arithmetic without creating a new observation family. Record a discrepancy without inventing its cause.

## Write a bounded evidence-use decision

What does the primary source establish? What extra work does each account contribute? What remains unknown? Which source should a memo cite for the number, and which for contextual reporting? What additional evidence would change the family assignment or resolve a discrepancy?

Before describing the claim as independently corroborated, check:

1. I can identify each source's relevant observations, rather than just its publisher
2. The claimed relationships have documentary evidence or are explicitly inferred
3. The compared claims have matching scope, period and population
4. Independent analysis has not been mistaken for independent data collection
5. Unresolved inputs and overlapping evidence remain visible
6. The conclusion makes no unsupported promise of accuracy or confidence

For origin tracing, see https://neweruditions.com/insights/trace-a-market-claim . For comparability, see https://neweruditions.com/insights/why-market-size-estimates-disagree .