Any tool can print a timestamp
Every short form analysis tool will hand you a list of moments. The trouble is that the list looks identical whether the tool actually read your video or simply produced timestamps that feel about right for something of that length. From the outside there is no way to tell, and that is true of ours as much as anyone else's.
That is an uncomfortable position for a product whose entire output is a set of moments. So instead of asserting that ours are real, we built a test that could have shown they were not, and we ran it.
The question we set out to answer
Retensis flags specific points on your timeline. The question worth answering is whether those points are anchored to the structure of the video, or whether the same list could have come from anywhere.
These are the moments that drive drop-off analysis, so whether they are real is not an academic question. If they are decoration, so is everything built on them.
How we checked it
We measured whether the moments Retensis flags are genuinely anchored to the structure of an edit, under controls fixed before the run.
That last part matters more than it sounds. A test designed after the results are in can always be made to pass, and a number produced that way is worth nothing. This one could have failed.
The complete figures are published in how we validate the moments Retensis flags, and the same sourcing rules apply to every other number in the product, as every number tells you where it came from sets out.
What the result shows
It cleared both controls, and not marginally.
What follows from that is specific and worth stating plainly: the moments Retensis flags are tied to the structure of your edit. They are not spread at random, and they are not generic points that would fit any video of that length.
The practical consequence is that a flagged moment is worth opening. It points at a real decision in your edit, so the timestamp is somewhere to look rather than something to take on trust. That is a smaller claim than most tools in this category make, and unlike most of them, it is one we can put figures and odds against.
The numbers
The test ran across 497 flagged moments and cleared both controls.
The odds of chance alone producing a result that good are about 1 in 900 against the first control and about 1 in 5,000 against the second. The run was produced by Retensis Vision 2 on 2026-09-19, and results are recorded per engine generation, so a later generation is measured on its own terms rather than inheriting these.
Frequently asked questions
We measured them, under controls fixed before the run, so the test could have failed. It did not. The odds of chance producing a result that good are about 1 in 900 against the first and about 1 in 5,000 against the second.
That something happens there worth looking at. The measurement shows these moments are tied to the structure of your edit rather than scattered across it, so the timestamp is a place to open and review rather than a number to accept on faith. It points you at a decision you made while cutting, which is the part you can change next time.
See what Retensis Vision finds in your next video
Upload a video or paste a YouTube URL and get a full multimodal breakdown of your hook, pacing, audio, delivery, and predicted retention, in about 90 seconds. Free to start.
Analyze your video free →