Three details a celebrity lookalike quiz does not print

Celebrity lookalike quizzes come in two kinds. A forced-choice quiz asks you to pick the closest celebrity from a few options. An upload tool such as StarByFace compares your photo with a celebrity library and prints a similarity percentage. Both leave out three details: the chance baseline, the photograph difficulty and prior exposure.

Detail one: the chance baseline

In a forced-choice quiz, the baseline is one divided by the number of options. Four options mean guessing alone scores 25 percent; twenty options mean 5 percent. On ten four-option items, guessing reaches four or more correct more than one time in five, so a 40 percent result proves little.

An upload tool has a different baseline: selection. It scores your face against every face in its library and shows the top few. Sites describing StarByFace say it ranks a celebrity library by similarity and shows a score out of 100. They also say the score measures closeness in the model's own numbers, not the probability of a real resemblance.

The highest of hundreds of scores always looks better than a typical one, and the printed percentage never says how many candidates it beat. Details of how face-match scores are calculated show why the same photo can score very differently on two tools.

A worked comparison shows the gap. Forty percent on ten four-option items is within reach of luck. Forty percent on ten twenty-option items would happen by guessing about once in a thousand tries. The two printed numbers match; their meaning does not. A quiz design check lists what else a score can hide.

Detail two: how hard the photographs are

A quiz that mixes well-lit press portraits with blurred candid shots measures the pictures as much as the player. Two quizzes can use the same faces and still give different scores, because difficulty sits in the image set. No printed percentage separates the two effects.

Upload tools have the same problem in reverse. The tool's score depends on your photograph: angle, lighting, expression and age of the picture all move it. Descriptions of StarByFace say as much. Two photos of one person can give different celebrity matches, which says something about the photos, not the face.

A fair quiz states what kind of photographs it used and says if clarity varied. A missed question on a hard image is not proof that a face is hard to recognize. The photo conditions behind a comparison and the ordinary image errors behind most bad matches explain the same effect.

Detail three: what the player has already seen

A score records responses to the images shown, not a stable skill. A player who has seen the quiz photographs before, or who knows a celebrity's famous publicity pictures, may do better than a player meeting them for the first time. The history never appears in the final percentage.

With an upload tool, the exposure belongs to the tool. Its library holds particular reference photos of each celebrity, and its model learned from face data that the page rarely lists. A match to an actor may reflect which of that actor's photos the library holds. See how lookalike quizzes train the eye for how earlier viewing shapes the result.

Compare people only when they answered the same pictures under the same conditions. Even then the result describes that image set, not general face memory. The accuracy limits of face-match tools apply to the upload kind as well.

What to look for before trusting a score

Before reading a celebrity match score, find three things on the page. How many options or candidates did the quiz choose from? What kind of photographs did it use, and were they the same for everyone? And whose pictures, yours or the tool's library, shaped the number? If the page does not say, treat the percentage as entertainment.