What Does a Quality Score Out of 100 Mean?
A file gets a number. 44 out of 100, say. What is that actually telling you, and what would it take for the number to be worth trusting?
The objection to a single number
Take a photograph that is perfectly exposed, free of noise, 24 megapixels, and completely out of focus.
Score each of those four things out of 100 and average them. Three near-perfect scores and one near-zero gives roughly 75, and 75 reads like a decent file. The photograph is unusable. Every copy of it is unusable. The one measurement that mattered has been diluted by three that did not.
That is the problem with averaging quality dimensions, and it is not a corner case — it is the normal case, because files usually fail in one specific way rather than degrading evenly.
A formula that cannot do that
The alternative is to let the weakest check set the ceiling:
- Score every dimension on its own scale, 0 to 100.
- Take the weakest one. That is the ceiling — a file is only as good as the worst thing in it.
- Move it down a little according to how much else is weak, so that one bad check and six bad checks are not the same answer.
- Clamp the result inside the band the weakest check earned.
The out-of-focus photograph now scores in the teens, because the focus check is the ceiling and nothing can lift it.
Why the clamp matters more than the formula
Step four is the part that makes the number trustworthy, and it is easy to skip.
A quality report shows a band — Excellent, Good, Fair, Poor — next to the number. If the number and the band can disagree, both become noise: a red 38 beside a green “Excellent” badge destroys confidence in the whole report, and the reader has no way to know which one to believe.
Clamping the score into the band’s own range makes disagreement impossible by construction rather than by care. The number stops being a second opinion and becomes the band read finely — where inside Fair, rather than is it Fair.
The property worth demanding
Ask one question of any quality score: can it go up when something gets worse?
If lowering any individual measurement can raise the total, the number is not measuring quality, and there will be a file somewhere that games it. A score built the way above cannot: lowering the weakest check lowers the ceiling, and lowering anything else increases the drag. Both directions point the same way.
That is worth checking rather than assuming. It is also worth checking exhaustively rather than on examples, because the drag term is exactly the kind of thing that breaks it in a narrow range nobody thought to try.
What the number is not
It is not a percentage of anything. 44 out of 100 does not mean the file is 44% as good as a perfect one, because there is no perfect file to measure against — that is the whole difficulty of judging one file on its own.
It is also not comparable across checks. A 70 for noise and a 70 for bitrate each mean “low Good on that check’s own scale”, and nothing more. They cannot be added, averaged or ranked against each other, which is precisely why the headline is a ceiling rather than a sum.
No original to compare against? DiffALL measures one file on its own and names what is wrong with it.
Check a file — free