Every Number in Retensis Tells You Where It Came From

Which numbers are the platform's and which are ours

Retensis keeps a record of where every published metric comes from, and there are three possible answers rather than two.

Some numbers are reported by the platform and passed through untouched. Views, likes, comments, subscriber counts. We do not adjust them, because the platform is the authority on its own figures.

Others are calculated by Retensis from data your connected channel returns. Subscribers per thousand views is ours. So is the share of viewers who stayed past your opening. These are useful precisely because the platform does not publish them, and every one of them says plainly that it is our arithmetic rather than the platform's.

The third kind is a reading taken from a screenshot you supplied, which is neither of the first two and is never quietly filed as one. The full breakdown lives in the documentation.

The number that looks broken and is not

Your average watch percentage can go above 100%. The first time you see it, it reads as a bug.

It is not. A section somebody watches twice is counted twice, so a video people replay can genuinely average more than a single full view per person. It is one of the more useful things a retention figure can tell you, because it points straight at the part of the video that survived a second viewing.

The easy thing would be to cap it at 100 and never field the question. That would also throw away the most interesting signal on the page, so instead the number is shown as measured and explained where it appears.

What happens before an analysis reaches you

Every completed analysis is checked against a fixed set of structural rules before it is saved. The rules cover what is wrong by definition rather than wrong by opinion: a timestamp falling outside the video it describes is an error no matter what the video is about.

The set is version locked, so it cannot be quietly relaxed when something inconvenient starts failing. Changing what counts as a valid analysis has to be deliberate and has to be reviewed, which is the only way a standard like this survives a deadline.

Why we score ourselves against real data

When your channel is connected, Retensis compares its own predictions against the retention the platform actually reported for that video. Not an estimate, and not a figure read off an uploaded image.

That distinction is the reason imported screenshots stay labelled as imports. They are worth showing you, and they are a different kind of fact from a measurement the platform returned, so the two are never averaged together.

It is the same discipline behind publishing the test we run on our own drop off detection, results and odds included, in how we test whether our drop-off detection is real. Claims we cannot support that way do not get published at all, which is the part that makes the published ones worth reading.