Video platforms adjust playback level so that one upload is not dramatically louder than the next. That adjustment changes what a loud master achieves.
Measurement is of perceived loudness
The measurement used is not peak level but an average weighted to approximate how human hearing responds across frequencies.
A track can peak at the maximum permitted value and still measure as quiet, or measure as loud without approaching the peak, because the two describe different things.
Platforms compute this figure for every upload and apply a gain adjustment so that playback lands near a common target.
Loud masters are turned down
A master compressed heavily to sound loud measures above the target, so the platform reduces its level on playback.
What remains after that reduction is the compression itself, meaning reduced dynamic range at the same perceived volume as everything else.
The competitive advantage that motivated heavy limiting therefore disappears, while its side effects on the sound do not.
Quiet masters are treated differently
Adjustment downward is straightforward, since reducing level introduces no new problems. Raising a quiet track is riskier because it can push peaks into distortion.
Platforms are consequently more conservative about increasing level than decreasing it, and some do not raise quiet content at all.
A master mixed well below the target may therefore simply play quieter than surrounding content, which is the one case where the system does not equalize.
Targets differ between destinations
Each platform sets its own target, and broadcast standards use different figures again, so one master cannot be optimal everywhere.
Mixing to the lowest relevant target and allowing louder platforms to reduce it is the approach that avoids distortion in every case.
This is the same reasoning behind broadcast delivery specifications, which state a loudness figure and a tolerance rather than a peak limit alone.
What this means for the mix
Since level is equalized regardless, the useful decisions concern dynamics, clarity and balance rather than how loud the file measures.
A mix with preserved transients sounds more impactful at a matched playback level than a flattened one, which reverses the incentive that produced heavy limiting.
Checking a mix at the platform's target level, rather than at whatever the master happens to be, is what makes that difference audible during the work.
Comparing two mixes at matched loudness rather than at their own levels is the same discipline, and it removes the bias that makes the louder version always sound better.