What Is a Review Queue?

The general idea

Any automated decision-making system, not just photo culling, can be built one of two ways. It can force a binary outcome on every case, which is simple but means low-confidence guesses look identical to high-confidence ones from the outside. Or it can expose its own uncertainty and set aside the cases it is unsure about for a person to handle. The second design costs a little more engineering, but it means mistakes concentrate in a visible, reviewable pile instead of scattering silently through the results.

Why a queue beats a forced yes or no

A review queue only works alongside some measure of confidence, since the tool needs a way to know which cases are uncertain in the first place. In culling, that measure is the confidence score attached to each decision: the queue is simply where everything below the uncertainty bar ends up. Without that signal, a review queue would either catch everything, which is no better than reviewing the whole shoot, or catch nothing, which defeats the point.

A concrete example

Picture a 5,000-frame wedding cull. Most frames are clear one way or the other: a black frame is a certain reject, a sharp, well-timed formal is a certain keep. A smaller set, maybe a few hundred frames, is genuinely arguable: a formal where one guest's eyes are half-closed, a portrait right at the edge of the sharpness threshold. Those are the frames a review queue is for. Instead of spot-checking 5,000 frames or trusting all of them blindly, you spend your attention on the few hundred where it actually changes the outcome.

Limitations

A review queue is only as good as the confidence estimate behind it. If the tool is confidently wrong, the mistake never reaches the queue at all, which is why spot-checking outside the queue still has value. And a queue only helps if someone actually works through it; a review queue that gets rubber-stamped without real attention provides little more than the illusion of oversight.

How Cull AI Studio handles this

Cull AI Studio's version of a review queue is the REVIEW folder. Every decision carries a confidence score, and anything below the uncertainty bar, 0.65 by default, is routed to REVIEW instead of being silently sorted into KEEP or REJECT. That keeps the photographer's judgment in the loop specifically where the software is least sure of itself, which is the practical answer to whether AI replaces the photographer's judgment: it does not, and the REVIEW queue is where that division of labor plays out on every single cull.