A similarity score is not a reunion probability

A score beside a photo result can look more precise than it is. Unless a system has explicitly validated and explained a probability interpretation, do not read a similarity score as “the chance this is my relative.” A ranking number can serve a narrower purpose.

Ask what the number measures

Some scores describe mathematical similarity between image representations. Others reflect an application's own ranking rules. Different tools may use different scales, and the same-looking number may mean different things in each system.

Wakif's face-search configuration describes development ranking, not a validated probability of identity. The number should not be turned into an accuracy or reunion claim.

Compare context, not just numbers

A higher-ranked candidate can still be wrong. A lower-ranked candidate may deserve review when contextual information is relevant. Check the report's date, location, and description, and retain any conflicts in your notes.

• What does the interface say the score represents?
• Is a probability interpretation actually supported?
• What relevant information was outside the comparison?
• What independent verification remains necessary?

Communicate the limit

When passing a lead onward, describe it as a possible match produced by the tool. Avoid rewriting a score as a percentage certainty in a message or poster.

No score should authorize confrontation, public naming, a handover, or a payment. The practical value of a ranking is to help organize review. Identity and safe next steps require an appropriate process beyond the number.