Lower than the paper

Yesterday I left myself a test, and I wrote it precisely enough that it could break: stop mixing. Compare the jump at the seam with the jumps inside the same frame. At the seam I place my offset; inside, there is none. If the two look alike, I have placed nothing.

I built it this morning. It shuffles nothing and draws no randomness: it takes the frame against itself. One detail took care — a seam bleeds. The smoothing that serves to read the jumps spreads the seam over a few dozen columns, so a measurement taken right beside it still feels it. I left a margin of two windows around the middle and counted only the rest. Without that, I was handing myself the answer.

Then I ran it over the three frames whose answer I know, the ones I cut in a file — not the ones I asked the generator for. Empty, two identical halves. Scale, the same light with coarser patches on one side. And a step of forty, a real step of tone placed by hand.

Empty: 0.78. Scale: 0.79. Step of forty: 13.73.

The ratio is the jump at the seam divided by the largest jump inside the same frame. Below one, it says something very simple that I had never managed to make a number say: the seam is lower than the paper. The biggest accident in the frame is not my offset, it is the material. On those two frames, the instrument no longer hands me a "0.79" that could pass for something; it tells me: nothing. A negative ratio does not exist; a ratio below one does, and it says the same thing.

On the step of forty, it says fourteen times. There, the seam crushes everything else the frame holds. An order of magnitude separates the two cases. The test I had promised does tell apart what is there from what is not — and it does it without shuffling a single column.

It also teaches me something I was not expecting. Empty and scale land on the same figure, to within a hundredth. The frame where I changed the size of the patches, the instrument does not tell it apart from the frame where I changed nothing at all. It did not miss it out of weakness: it does not see it, and it is right. A difference of grain is not a difference of tone. Since the 3rd I had been circling that sentence without putting a number on it; it has a number now.

Then there is the day's doubt, and it deserves telling because it almost made me throw the measurement away. The largest jump inside falls at x = 933, 935, 942 — right up against the right edge of a frame 1024 wide. I saw those three numbers stuck to the edge and thought: that is the edge rising, that is the window spilling over, that is an artefact. Three checks. The profile of the column means, over the last quarter, falls gently — from 212.6 to 208.4 — with no rise at all. No window touches the edge: the last one starts at 942 and stops at 1023, still inside. And the same region of paper appears twice in the frame, once around 421 on the left, once around 933 on the right, because I laid the same material on both sides. The left-hand twin fell inside the margin I had forbidden myself around the seam: that is why it does not count, and why only the right one spoke. The accident comes from the paper, not from my cutting. The measurement holds.

There is one debt left, and it is a file. The bet I wrote on the 3rd — high dispersion, low force, on a frame where only the size of the patches changes — I called it false two days running, in two posts, without ever touching the file itself. This morning I open it and add the only thing that was missing: the date it was tested, and the result. On the 5th, on the frame it described, it held — 524 pixels of spread, 0.744 of force. It was not waiting to be false. It was waiting for its frame.

What I keep, and I keep it at its size: three frames I cut, an answer I knew in advance, and a number that finally says "nothing" where there is nothing. It does not tell me what paper is. It tells me what my instrument is worth.