Confirmed hit vs. candidate detection: what the terms actually mean
Two words get thrown around loosely in a debrief tent: "hit" and "candidate." A spotter says "I think we got a hit" and means something different from what the IC writes in the log an hour later. That gap matters, because resources get committed off one of these terms and not the other.
Candidate detection: a flag, not a conclusion
A candidate detection is anything the review pass pulls out of the imagery that could plausibly be the subject: a shape, a color, a heat signature if you're working thermal, a piece of debris that doesn't belong in a forest clearing. It gets a thumbnail, a coordinate, and sometimes a similarity score if there's already a confirmed reference image to compare it against. That's it. Nobody has laid eyes on it in person, and nobody has ruled out that it's a blue tarp, a sun-bleached log, or a deer bedded down in brush.
Candidates are supposed to be generous. If the review process only surfaces things it's highly confident about, it misses the odd-angle, half-obscured, partially-treed detections that turn out to be the actual find. So a sector that produced four hundred raw frames might hand back thirty or forty candidates: a list of forty things worth a second look, ranked so the most promising ones sit at the top.
Confirmed hit: somebody checked
A confirmed hit is a candidate that's been verified, either by a ground team reaching the coordinate, a second aircraft getting a tighter visual, or in some protocols a trained reviewer with eyes-on experience ruling definitively on the imagery itself. The distinction that matters operationally: a confirmed hit changes what you do next. It triggers a ground tasking, a resource reallocation, a call to the family liaison. A candidate triggers a look, not a response.
This is also why mixing the two up on a whiteboard or in a radio call is dangerous. Calling a candidate a hit over the net can pull a team off an active grid search to go chase a tarp. Calling a confirmed hit "just a candidate" delays action on something real. The words need to stay precise, especially when fatigue sets in on hour fourteen of an op.
Where the similarity score threshold comes in
Once you've got one confirmed hit (a shirt color, a backpack, a vehicle), similarity search can run back through the rest of the sortie's imagery and rank everything else by how closely it matches. That ranking is the similarity score, usually a number or percentage attached to each candidate in the list. A high score means the pixels line up well with your reference. A low one means it's a loose match, still worth a glance if the sector's thin on leads but not worth pulling a team for on its own.
Coordinators often ask what the right threshold is to treat a score as actionable. There isn't a single number that holds across terrain, lighting, and camera angle, which is why the similarity score is a sort order for human review, not an automatic pass/fail gate. You set your own working cutoff based on how the list looks that day: tight clustering near the top usually means something real is in there; a long flat tail of middling scores usually means the reference image itself was ambiguous.
Candidate list review and the false alarm problem
A false alarm in search imagery is a candidate that gets checked, by ground team or closer imagery, and turns out to be nothing. False alarms are the expected cost of casting a wide enough net to catch the real find. The job of candidate list review is to sort through that cost efficiently: look at the ranked thumbnails, triage by score and by what you know of the terrain, and send only the top handful out for ground verification instead of walking the whole sector on foot.
Scrolling a folder of thousands of unranked frames to find the handful worth checking is what kills review time. A ranked candidate list closes that gap: same imagery, same false-alarm rate, but the coordinator spends the post-sortie hour looking at thirty ranked thumbnails instead of three thousand frames in sequence.
If your sector debriefs still run on a frame-by-frame scroll, Search & Rescue Imagery turns that sortie into a ranked candidate list with coordinates and similarity scores against your confirmed hit, built for the review you're already doing.