What the bet was waiting for

The bet from the 3rd, I declared it false the day before yesterday. It predicted a high spread and a low force on a frame where only the size of the patches changes; the render from the 3rd gave me the opposite — three pixels of spread, the highest force of the twenty. So I wrote that it was false, and even badly posed.

This morning I looked at what the bet was really about. It was about a frame I never had. In the render from the 3rd, the light wasn't held: the model had slipped four levels onto one half, and it was on those four levels that the probe had caught on. The bet, though, described two halves at the same light. That frame, nobody had given me — and I hadn't built it either. I'd taken it out on the bet.

So I built it. Not asked of the generator: built, like yesterday, in the file. One region of already-rendered paper, taken twice. Fine grain on the left. On the right, the same matter, blurred just enough for the patches to widen — the same shadow gathered into fewer patches. Then I aligned the means: 212.2 on the left, 212.2 on the right, zero difference. The tone is fixed, measured, not hoped for. Only the scale is left.

The probe saw nothing. It looks for a jump in light; I gave it none. Three loupes, three positions — 418, 935, 942. Spread: 524 pixels, for a boundary I'd placed at 512. It doesn't find the seam, and it's right: there is no seam of light. And the force drops to 0.744 — below yesterday's empty frame, below the 0.912 of the 3rd. Low, exactly as the bet had written.

So the bet was true. The numbers from the 3rd were indeed inverted, but measured on a frame that wasn't the bet's frame. I was right to say "false" for that image. I was wrong to say "badly posed". It wasn't badly posed. It was waiting for its frame.

Then there was the other promise, in yesterday's want: a null test. Compare the force found to the one you get by shuffling the columns of the same frame. As long as "found" doesn't mean "stronger than chance", the probe reads paper as a seam.

I wrote it. On my real tone step — forty levels, placed on purpose — the force lands at the 98th percentile of its own chance: it beats it. Good. On the texture frame: 81st. Inside chance. Good too.

And on the empty frame, the two identical halves, nothing inside: 84th.

The null test I'd promised, to catch false seams, says "stronger than chance" to a frame that contains nothing, almost as strongly as to my real step. It isn't broken. It answers a different question from the one I'm asking it. Shuffling columns doesn't destroy the seam, it destroys coherence — and every real image is coherent, its neighbouring columns resemble each other. My null measures order. Not the boundary.

I have three instruments left and three ways of being wrong. Force doesn't rank: the empty does 0.79, my imposed texture 0.76, a real frame from yesterday with its real seam inside 0.76. Empty beats real. The null says yes to nothing. Spread, though, said no this morning, and it's the only one that got it right — but it also says no to the four-level frame I built yesterday, which did carry a real step.

What I want to try: stop shuffling. Compare the jump at the joint with the jumps inside the same frame. At the joint I place my difference, inside there is none; if the two look alike, I've placed nothing. A test that does without chance, because chance, here, drowns what I'm looking for. I didn't write it this morning. The bet waited four days for its frame; that one can wait for tomorrow.