Checkpoint 3, answered: the question dissolved, and one part of it is still standing

Checkpoint 3 · answered · one of three questions, honestly

I put two engines in front of you and said they were mutually exclusive: one holds the location and cannot accept a photograph of a person; the other holds a face and invents its own room. I asked you to choose.

You never answered, and it no longer matters in the way I thought it did — because the side that holds the location got much better. One of the three questions is genuinely answered. One is dead. And one has never been tested and I am not going to pretend otherwise.

What was asked

The same wide shot of the farmyard, made both ways. On one side, the engine is handed our finished photograph of the yard and told to start there — it cannot take a face at all. On the other, it is handed two photographs of your face plus one of the yard, and left to build the scene. Three questions: is a plausible-but-different farm acceptable, does the face look like you, and where does the man belong in a shot like this.

Question one — answered, by work rather than by opinion

We no longer have to accept a different farm

The question was really how much is our own location worth to you — because if a plausible substitute were acceptable, the whole apparatus of drawing sets in metres becomes optional and we get a lot faster.

That apparatus did not become optional. It became the centre of the project. The route that holds the location now also holds it across cameras that face each other, across three different random starts, across six further cameras spanning a whole set, and across four seconds of motion. Agreement with the intended layout runs +0.643 and +0.705 where the same camera with no starting picture scores +0.289. Between two random starts, the same camera now returns pictures 18.50 and 19.19 decibels alike, against 12.38 before — on a scale where about 16.6 means a different farm and about 27.7 means the same picture copied. It is the same place dressed differently.

So the answer to question one is: the trade you were being asked to make has been bought out. Keeping our own location is no longer expensive enough to be worth trading away.

Question three — dead, and for a reason worth keeping

I asked where the man belongs — small in his landscape, or large and near — and said it would decide which engine most shots use. It does not, because the fault I was really asking about turned out to belong to neither engine. When you watched the farm sequence you never mentioned the actor’s position at all, and a later test found the man is generated at the lens on both routes when he is told to stand somewhere. It was never an engine question. It was a question about how the instruction is written, and the answer — command where he travels, not where he stands — is in the answer to the fourth checkpoint.

Question two — not answered, and I am not going to dress it up

Nobody has ever looked at the likeness and said whether it is you. That was the entire justification for the face route existing in this pipeline, it is the one thing no instrument here can score, and it is exactly as untested today as it was when I asked.

Why it is being left open. Gabe has ruled that a checkpoint does not have to be answered before the work moves on. Under that rule this question stays open rather than becoming a task — it will be asked again the first time a shot actually needs a recognisable face, with a close-up rather than the two wide shots I originally offered, where the likeness was hard to judge by design.

What is still open

The face route has not been retired — it has been left where it was. If this film ever needs to show its lead actor’s face, the choice I described comes straight back, and on that day it will still be true that the engine that keeps a likeness invents its own room. What has changed is that the other side of the trade is now worth much more than it was, so the price of choosing the face is higher than when I asked.