Diagnosing a key reading you don't believe
The readout says F minor. You've played the song twice since it came up, it still says F minor, and you still hear A flat.
One of you is wrong about where home is, and it probably isn't the part of the software that reads pitches.
Pitch detection is the solved part of this. A detector listens for which frequencies are present and how long they're held, and on an ordinary record it gets that inventory close to right. The hard part is the step after, where it decides which of those pitches the music treats as home. That decision isn't in the notes. It's in phrasing, and in what the bass does at the end of eight bars, and plenty of records never settle it.
Six situations produce a reading you'll argue with. You can tell them apart from the outside, without knowing anything about how the software works.
Count the distance first
If the answer is three semitones from the one you expected, in either direction, stop looking for a bug. A minor and C major contain the same seven notes. So do F minor and A flat, which is the pair at the top of this page. Nothing in the pitch content separates a relative pair, because there is nothing there to separate, the two keys being the same collection of notes with a different one treated as home. A detector that picks the wrong half of a relative pair read the record correctly and then guessed at the one thing the record left open.
It read the intro
An eight bar pad intro is, to a detector, eight bars of evidence. If that pad sits on the four chord the whole way, the four chord gets the weight.
Run the analysis on the second chorus instead, and see whether the answer moves.
The song doesn't have one key
Some records modulate. Others borrow a chord from somewhere else and come back, which isn't a modulation but will still pull an analysis window toward a different center for as long as it lasts. Either way, a single key printed next to a filename is a summary, and summaries lose things. If the reading you got matches the bridge and not the chorus, you're looking at a file that doesn't have one answer to give.
The material isn't major or minor
A two chord Dorian vamp never cadences. Nothing in it resolves, so nothing in it declares a tonic, and asking whether it's major or minor forces a choice the music declined to make. Modal material and a lot of loop based production sit here. What you get back is the closest available label, and the closest available label can fit badly without being wrong.
The third is the quietest note in the mix
One note separates major from minor, and it's the third. In a lot of modern mixes the third is the least present harmonic information in the whole record, sitting under a sub or implied by a vocal line without ever being played by anything. When that's the case the detector has almost no evidence for the exact thing it's being asked to decide, and it will still hand back an answer that looks identical to one it had good reason for.
Session™ Key puts that difference on the reading. It shows green only once it has committed, and it stops listening at that point so the answer holds. Anything short of that stays amber, with the candidates it is choosing between named on the panel, which is where a drum loop lands because there is nothing in a drum loop to be sure about.
The record isn't at concert pitch
A sample pulled off a record that was mastered slightly fast sits between two keys. So does a synth nobody tuned to 440. A detector rounds to whichever key is nearer, and it can round differently in different sections of the same file.