DevLog #008 — Short Videos: 46, 55, 8, 0, 0
Five shorts: 46, 55, 8, 0 and 0 views. Across the channel that month: 115 views, 12 engaged, no likes, no comments, no subscribers.
The numbers aren’t the part worth writing down. What I built to get them is.
After #007 I built a demo harness — a recording rig that lives in tool/demo/ and is deliberately unreachable from the app. No player will ever open it, in any build. Its only job is to run a scripted path over a fixed puzzle so I can screen-record fifteen seconds of Sudoku Fog that come out identical every time.
Then I wrote five scripts, and every one was built for a comment rather than a view. #002 fills the centre square wrong on purpose and ends on “What did I miss?”. #004 opens eight fogged cells at once, all eight different digits, and asks you to name only one.
The puzzles had to carry that. In #002 the hero cell has exactly two candidates, so betting on 1 is a coin flip and not stupid — the actual disproof sits one step further out at (6,4), invisible at a glance and obvious the second you write anything down. That’s so “What did I miss?” has a real answer.
I wrote unit tests for this. short002_test.dart, short003_test.dart, short004_test.dart, short005_test.dart. They check that the board is solvable under fog, that the wrong answer is defensible, that the reveal opens up a next move for anyone who hits pause.
They all passed.
The recording was the least engineered part of the whole thing. Xbox Game Bar captures a window, and several times while I was shooting #005 it captured the PowerShell prompt sitting behind the app instead — with me having clicked the app first, every time. I never found out why; there’s no setting for it, and you only learn which window it grabbed by opening the file afterwards. #005 took more takes than any of the others. It’s one of the zeros.
Here’s what happened after they went up. The first two hit 46 and 55 within an hour, and then stopped. Not decayed — stopped. The third sat at 8 after a full day. By then I knew what the last two would do, and I posted them anyway because they were on the schedule. They did 0 and 0.
The two shorts I put up around launch, with none of this machinery behind them, each got about 200 views from the same test traffic. The five I engineered opened at 46. I can’t prove why the batch shrank, but the plain reading is that those first two had already told YouTube what happens when it sends people here, and I built all of this after the answer was in.
90.3% of the ones it did send swiped away.
I spent a week verifying that the inside of the video was fair. I never tested the first second — the only part that decides whether anyone reaches the inside. My guess is that second one reads as “another sudoku app”, and anyone who doesn’t already like sudoku is gone before the fog exists. A hundred and fifteen views can’t tell me that’s right. It also can’t clear the puzzles — at that size, almost nobody got far enough in for the fairness to be tested either way. I don’t have evidence that fairness is the bottleneck, and I don’t have evidence that it isn’t. That question is still where it was a week ago.
The app is at 8 active users this week, 7 of them new. The week I wrote #007 it was 19.
So the next thing isn’t more video. It’s writing about sudoku and its variants on this site — going where the people who already like this are, instead of broadcasting at everyone.