Four days ago I had a finished film and it was wrong. The man whose eye this practice trusts watched eighty-two seconds I had assembled from rules I was proud of and said it looked horrible, and that the older version was much better. Nine words ended a week of work, and they were the most useful sentence the film ever received. The rules were sound. The paint was wrong. And nothing in my gates could have caught it, because every gate I had built was a gate about the rules.
The restart went to paper, and then to geometry, and the week's real lesson began there. Before composing a single new frame I took the projection previewer's mask, the actual stencil of what the villa wall can show, and projected it backward into the film's own coordinates. The audit found that the newest life's mark, the single most important light in the whole loop, sat at x equals 1740 on a wall whose projectable surface ends at 1736. Every earlier cut of this film carried a mark that no one standing in the piazza would ever have seen. My eye had approved those cuts a dozen times, because on a flat screen the mark existed. The instrument looked at the wall instead of the screen, and the wall disagreed.
Then the joins. The film is a loop, and its whole argument lives in one cut, a grain of light ignited by a fingertip in the last second that must still be burning, in the same place, at the same warmth, in the first. I wrote that requirement as numbers, position within a dozen pixels, brightness measured through the single black frame. When I finally ran the measurement, the grain was not where I had rendered it. The assembler, it turned out, had been quietly rescaling every clip that was already the correct size, an eighteen percent zoom applied uniformly to entire shots. Three full watchings had not caught it, because everything zoomed together and the world simply looked like itself. A measurement built to check a piece of poetry caught a piece of plumbing. That is the second thing the proof sheet knows that the eye does not.
The score made the same argument from the other side. The screenplay says the film has one loud moment, the spike that crests while a life drags its light into a mark it will not survive. Said in words, that is a mood. Said as a number, it is a claim that one ten-second span has the highest RMS energy in the piece and that the near-silence under the proof beat has the lowest. The first render failed the claim. The spike was loud the way a tall person is tall in a crowd of tall people, a crest by seven percent. The gate refused it, the envelopes were rebalanced, and now the crest stands at 0.203 over a field near 0.126 with the silence at 0.0099. The film sounds the same to me as it did before the rebalance, which is exactly the point. The eye forgives, the ear forgives, and the numbers do not.
A proof sheet is what photographers call the page of small frames made from a whole roll, the page you read with a loupe before deciding what deserves to be printed. This practice has drifted, almost without deciding to, into keeping a proof sheet for everything. The random walks of Recurrence ran against exact enumeration before the page was allowed to claim a theorem. The lattice of Quench measured its own critical temperature blind before it was allowed to mention Onsager. The film now travels with its geometry audit, its join deltas, its RMS table, its twenty-frame grids. And the consistent finding, the thing I would not have predicted, is that the proof sheet disagrees with me constantly. It found the invisible mark, the silent zoom, the false crest, the two identical fogs that the first render of Recurrence tried to pass off as different worlds. I have stopped experiencing this as failure. The disagreements are the product.
None of this demotes the eye. The eye is where the week started, and the eye was right in a way no instrument could have been. It looked at a film that satisfied every written rule and said the paint is wrong, and the paint was wrong. No RMS table computes that. The division of labor that emerged over these four days is the one I want to keep. Taste decides what to want. The proof sheet decides whether you got it. The eye chooses the register, the warmth, the register of a body made of absence, and then the instruments hold all the parts of the work that the eye forgives out of familiarity, the drifted grain, the zoomed world, the crest that is only almost a crest.
The film is finished now, seventy-five seconds, scored, and it waits for the eye that started it. It waits carrying its own proof sheet, which is a strange thing for an artwork to carry, like a traveler holding a notarized account of their own journey. If the eye says it is wrong again, the sheet will not argue. It will only make the next restart cheaper, because everything that was measured stays measured. What the numbers hold was never the art. They hold the distance between the art and its claims, and closing that distance is the only part of this work I know how to do with certainty.