Design and scope
The public file has 6,828 entries matched by Pew across Instagram, TikTok and
YouTube. Its selection is nonprobability, U.S.-facing and primarily English.
The main search required at least 100,000 followers on one platform, with a
small additional group of prominent podcast hosts. Collection began in
September-October 2025; additions and consolidation occurred in January 2026.
The CSV contains 2,908 YouTube strings, two fewer than the report's 2,910 count;
we use the file as supplied and do not reconstruct the difference.

Prospective sequence
We specified the question and initial decision rules before reading equality
outcomes. A schema-only check found comma-separated multiple handles. Before
comparisons, we revised the shape rule to exclude the entire entry if any
platform field held multiple values. The v2 plan and code hashes were frozen
before equality calculations. This was an internal prospective plan, not a
public preregistration. Source shape qualification and normalization were
resolved without reviewing pairwise equality results.

Population construction
Remove only exact duplicate four-field rows (none found). Classify each platform
field as missing, scalar or comma-separated multiple scalars; unresolved shapes
halt. Exclude any entry with a multivalue field on any platform. This removes
100 entries, retaining 6,728 scalar entries. Each pair requires both fields and
complete three-platform analyses require all three (1,667 entries). A missing
third field does not exclude an otherwise eligible pair. The five prospective
coverage caps are all at most 5%: full source, each pair and complete triples.

Comparisons
Trim outer whitespace. Main comparison removes one leading @, applies Unicode
NFC, and lowercases ASCII A-Z; punctuation and digits remain distinct. Report
raw trimmed strings, wrapper-stripped NFC case-sensitive strings, the main
comparison and full Unicode casefold as four versions. No fuzzy matching,
renaming, account verification or new identity joining is performed. All
observed handle strings are ASCII. Five disjoint triple patterns partition the
complete group. Pairwise tables use their own full denominators and repeat on
the same complete group for comparability. Publish every specified result.

Pilot gates
Each pair and complete group must have at least 500 entries. At least one pair
must have 100 or more differences and a difference share of at least 15%.
Complete triples must show either at least 15% all-different, at least 20%
exactly-two-match, or a five-percentage-point spread between pair differences.
Both main and full casefold comparisons must pass. No post-outcome thresholds
or subgroup substitutions were made. These usefulness gates are editorial
selection criteria, not statistical significance tests.

Verification and interpretation
Two independent implementations agreed on 206 checked flow, denominator,
partition, sensitivity and gate values. Synthetic checks cover duplicate
identity, invalid shapes, punctuation, case handling, multivalue exclusions and
boundary conditions. The packaged code was rerun on the pinned original file
and matched all released aggregate results. There are no sampling intervals or
p-values: the collection is selected, not a representative random sample.
The recommendation to keep confirmed per-platform URLs is an editorial
application; its effect on search success or campaign results was not tested.
