Evidence-backed creator intelligence
Paste a channel. We pull their real numbers, find the videos that quietly beat their own median, and hand you ranked ideas with the receipts still stapled on.
For a full channel pull
Each one with citations
Everything hangs off it
How it works
01 / Collect
Two to six minutes, running in the background. You get a live status card in under a second, then you are free to go and do literally anything else.
02 / Compare
Not a global benchmark, and definitely not the mean — one runaway video hauls a mean up high enough to hide every other outlier behind it. The median does not flinch.
03 / Commit
Each one cites the videos it came from and wears the confidence the model actually reported. Especially the embarrassing ones.
The non-obvious bit
On a channel averaging two million, it is a disappointment. On one averaging twenty thousand, it is the best thing that ever happened to them. Absolute view counts tell you almost nothing.
So every video is scored against its own channel's median. That one decision is most of the accuracy in this product.
Absolute view counts are meaningless. Relative ones are the product.
What a channel actually looks like
Every point is a video, placed by how far it beat or missed the channel's median. The crowded middle is that channel on an ordinary Tuesday. The stragglers at the edge are why you are here.
Generated ideas
Illustrative. Real ideas cite the videos they came from.
Receipts included
Every idea carries the confidence the model actually reported. A 41% stays 41%. Rounding it up to something reassuring would make the number worthless, and you would stop trusting the 82% too.
What is in the box
Every video measured against its own channel's midpoint, so a big number on a big channel stops flattering itself.
Context from beyond YouTube, so an idea turns up already knowing what the rest of the internet said about it.
The model's own number, unrounded. A 41% stays a 41% rather than being quietly promoted to a recommendation.
Slow work goes to a queue. Close the tab, make tea, come back — the scrape got on with it.
A million views means nothing without a baseline. Every video is measured against its own channel's median, so a big channel's flop cannot masquerade as a hit.
Each idea links the videos behind it, so you can argue with it properly instead of just vibing.
Deliberately out of scope
A tool that claims to do everything does nothing in particular. These are refusals, not a roadmap with the dates rubbed off.
Reasonable questions
No. Everything used is public: the videos, the view counts, the upload dates. It is the same information anyone can read on the channel page, gathered and compared properly.
One breakout video pulls an average up far enough to hide every other outlier on the channel. The median sits in the middle regardless, so a video that beats it has genuinely beaten that channel's normal.
Two to six minutes for a full pull, running in the background so you can close the tab. If a scrape returns 480 of 500 videos that is a success — the dropped ones are counted and logged, not silently ignored.
It stays in your own project. Row-level security is enabled on every table in the same migration that creates it, so the database itself refuses to hand your rows to anyone else.
Three free runs a month is more than enough to catch us bluffing.