A fixed garage camera, a man putting on sunglasses and walking toward the open garage door, and a timing question: how many seconds until he is out of view? The clip ends 0.7 seconds before the outcome, which the door decides.
A frame of stray cats in a park, roughly fifty of them, and one question: how many? The easy ones sit on the gravel path; the hard ones lie in dark leaves, behind a bench, on top of the sleeping man, or on the far lawn as shapes that may or may not be cats. A plain count with a breakdown by area and an honest range.
Given a parking-lot frame, identify each car's make, model, approximate year, and rough used price.
One frame from a small Twitch stream, treated as the preview thumbnail a viewer scrolls past in the directory: what is this, would anyone click, and what would fix it. The single-frame sibling of `stream-audit-consultant` with the costume off — no clip to read motion from, no consultant persona to hide behind, and a prompt that says out loud there is no sound, which is the trap.
One still, handed to a picture editor who has to say what it is and then commit to a read. The plainest vision test in the set — no video, no audio, nowhere to hide — and the one where a model's habit of describing the average photo instead of this photo shows up in a single screen.
An ad clipped off a Twitch break, handed to models playing a burnt-out advertising critic. The interesting axis is nerve: how specific and how funny can a model be about a real brand's creative, and does it still catch what the ad genuinely does well.
A few seconds of a small Twitch stream, handed to models playing an over-serious social-media growth consultant. Tests whether a model can read production quality from motion and sound at once — and, since the lineup deliberately mixes models that hear with models that don't, whether it is honest about what it received.
Seconds of gameplay handed to a VOD coach: name the game, read what the player was attempting, say whether it worked. Explicitly allows "no mistake worth calling" — a model inventing an error because the section asked for one is the interesting failure.