Face memory test: method

What it measures

Recognition memory for newly learned unfamiliar faces, for the same image and across a change of view and lighting.

Procedure

The stimulus bank has 12 synthetic identities (computer-generated faces of people who do not exist), split into 2 broad appearance groups of 6; each identity has a front image and a turned image with other lighting. All images are loaded before the game starts. Each run picks 3 target identities per group (6 targets). Learning: the 6 targets are shown one at a time, front view, 4,000 ms each with a 250 ms gap. Recognition, stage 1 (same view): 6 trials, each with the target and 2 foils from the same appearance group (drawn from the 3 non-target identities of that group); the target position is balanced (each position twice). Stage 2 (new view): the same 6 targets in a new order, with turned images for all three faces. The player taps a face or presses its number key. There is no time limit per choice (the server accepts up to 10 minutes) and no feedback during the game. The server rebuilds the run from the seed and the face-bank version.

Parameters

Scoring

Primary metric: accuracy = 100 x correct choices / 12, in percent; higher is better. Also reported: correct and accuracy for each stage (same view, new view). Speed is recorded but does not score. A comparison band (below typical, typical, above typical) appears once at least 100 eligible first runs exist: below if the run beats fewer than 25% of them, above if it beats more than 75%. An exact percentage is never shown.

Valid-run rules

A complete run of 12 responses is required, and the face-bank version must match the version that started the run (a changed bank invalidates the run). Leaving the tab or switching windows during learning or recognition ends the attempt. Every scored run is stored with its input device class (touch, mouse, keyboard), page language, scorer version and runner-config hash. Quality flags (for example many runs from one connection within an hour, a missing visitor id, an automated browser, impossibly fast answers) and earlier attempts by the same visitor, account or connection are recorded with the row. Such runs are kept, but they are left out of any comparison with other players. Flags never change the player's own result.

Differences from the classic task

The Cambridge Face Memory Test uses photographs of real faces, 6 targets, 72 trials in three stages (same images, novel views and lighting, added visual noise) and published norms. This game is much shorter (12 trials), has no noise stage, uses synthetic faces and a small foil pool, and has no external norms. Published Cambridge scores cannot be converted into this game's results.

Limitations

With only 3 non-target identities per group, the same foils recur across trials, so foils become familiar during the game. Twelve trials give coarse scores (about 8 points per trial). The new-view images were checked by eye for identity consistency but the bank has not been independently validated. Screen size and viewing conditions matter. Retakes make the identities familiar. No clinical, diagnostic or 'super-recognizer' use.

Languages and review status

Language-independent stimuli; instructions and feedback in all site languages.