Scoring explained

Candidates are scored from 0 to 100. The default model is deterministic and explainable.

Signal	Weight	Purpose
Handle similarity	20	Catches suffixes, one-character edits, and homoglyph-style handles.
Display-name similarity	15	Catches copied profile names.
Bio similarity	10	Catches "official", support, backup, and protected-handle references.
Image similarity	25	Uses perceptual hashing when profile image URLs or fixture hashes exist.
Account age	10	Fresh accounts are more suspicious.
Follower ratio	5	Tiny follower counts against a larger protected account are suspicious.
Follow-back pattern	5	Accounts targeting the protected audience get extra weight.
Posting behavior	5	Scam posts and protected-handle mentions raise the score.
Verified status	5	Paid verification can make clones more dangerous.

Mitigations reduce scores for parody/fan disclaimers, verified affiliation mismatches, and accounts that predate the protected identity.

For a narrative explanation of how clone tactics change over time, see How impersonators evolve.

Demo examples

• @alex_charts1: critical. Similar handle, identical profile image, new account, scam posts.
• @alex_charts_giveaway: medium. Suspicious promo account, but weaker image match.
• @alex_charts_fan: filtered. Explicit fan-account disclaimer.
• @alex_charts_real: filtered. Older unrelated account.