You do not need any of this to use Stellody. It is here because the claims on the Why page are the kind that are easy to make and hard to check, so this is the working behind them. The code is public and you are welcome to go and look.
The rules on the Why page are enforced by tests that stop the checks passing. Each of those tests was itself checked by deliberately breaking the rule and watching it fail, because a guard nobody has seen fail is not yet a guard.
In the collection Stellody was built against, 162 of 617 albums come from a cue sheet rather than one file per track, 155 of them a single file holding the whole album. That is a main path rather than an edge case, so it is captured once in the core idea instead of special-cased all over the program.
A full text index was the first plan and measurement refused it: reading the whole library on every keystroke costs a millisecond or two, against the hundred and twenty a typed character allows. An index would also have held the wrong text.
Grouping albums by their tags was tried against a real collection and failed: it split one Mozart folder into five albums, two holding a single track. A folder boundary is what a ripper actually recorded, so that is what is trusted.
Working out the shape drawn along the bottom of the window used to take over three seconds for a long track. Doing the same arithmetic a different way brought it under one; the picture that comes out is identical. That is how things get decided here: time it, then keep the quicker way only where it gives exactly the same answer.
Formats differ in what they will tell you: a FLAC states its bit depth, an MP3 states none. Where a file says nothing, Stellody records that it said nothing rather than filling in a likely number, because a made up figure is impossible to tell later from one the file really carried. That is what lets it refuse to claim a lossy track is bit for bit perfect, whatever it is played through.
The part of Stellody that decides what counts as an album, which tracks belong to it and what it should be called never reaches for your disk, the internet or even the clock.
It is handed what it needs and works from that alone, which means it can be tested completely rather than nearly. It is: every one of those decisions is put through a test each time the checks run, so it behaves the same on your machine as on anybody else's.