このガイドの全文は現在、英語のみで提供されています。
From hits to attribution
A basic scan produces candidates. Attribution is the separate discipline of deciding which candidates are the same human, and it rests on evidence the subject published rather than on string similarity.
The minimum defensible standard is two independent links: a reused image and a self-declared cross-reference, for example, or a shared unique bio phrase plus a linked personal page.
Permutations done deliberately
Rather than guessing wildly, derive variants from a rule set: separator substitutions, trailing year ranges around a plausible birth year, common prefixes and suffixes, and the same stem with a platform name appended.
Each variant costs a request, so prioritise. Test the small set you can justify, and record which strings you tried — a negative result is only meaningful if you can say what was excluded.
Archived sources and timestamps
Archived captures let you see a profile as it was, which is often more informative than the live version: follower counts from three years ago, an old display name, a linked site that has since moved.
Timestamps are the other quiet source. Posting times across several accounts reveal a consistent timezone and a sleep window, which supports or undermines the hypothesis that two accounts belong to one person.
Stylometric de-confliction, used carefully
Writing style — punctuation habits, sentence length, recurring phrases, emoji choices — can distinguish two people using the same handle, or suggest one person behind two accounts.
Treat it as supporting evidence at best. Style analysis produces confident-sounding conclusions from thin material, and alone it should never carry an attribution.
Writing it up defensibly
Keep two columns: what you observed, with the URL and the date, and what you infer from it. Every conclusion in the second column should point at a line in the first.
That habit is the difference between a report and an accusation. It also makes your work checkable by someone who does not share your assumptions.