The light they built for

Same rule as the earlier parts: the photographs first, and the arithmetic afterwards. Each pair below is the identical camera — same position, same bearing, same aim point — separated only by the hour. Golden at 0.64, then twilight at 0.71.

Every exterior in this gallery had been shot at the same hour. Not by accident — a measurement in the first series said golden hour beat the alternatives, and after that nobody looked again. Two thousand photographs later that single unexamined constant was the largest thing in the pipeline nobody had tested.

What restarted the question was a coincidence in the data. The three highest-scoring frames in the entire corpus all sat at 0.71 — an hour the exterior rig had never once used. All three were interiors, shot through a window by a different planner that happened to schedule dusk.

To place a camera relative to the light you have to know where the light is, and the mod does not report it. It turns out the photographs do. Every frame already records the bearing its camera faced; the sky at the top of the frame is brighter toward the sun. Regress one against the other and the sun falls out.

The first attempt failed, usefully. Averaged over the whole frame the luminance is dominated by the building — a dark hall and a pale one differ far more than two skies do — and the fit came back at R² near zero. Restricting the measurement to the top eighteen percent of the frame, and subtracting each build's own mean so the subject cancels, recovers the signal.

Seven independent capture runs, all shot at 0.64, agree on a bearing in the south-west:

Good to about ±25°. Enough to put a camera on the far side of a building from the sun; not enough to place a sunset behind a particular chimney.

With the bearing known, every golden-hour frame in the gallery could be labelled by how it stood to the light. Back-lit, side-lit, front-lit — a distinction that every photography guide treats as fundamental.

Nothing. A spread of three hundredths across 1,087 frames, against a build-to-build spread of 0.275. Either lighting direction genuinely does not matter in Valheim's soft golden light, or the thing doing the judging cannot see it.

It is the second one, and finding that out was worth more than the shot it was meant to plan.

The gallery ranks on an aesthetic model borrowed from open image research. Asked what moves its score across 2,181 frames, it gives a very clear answer, and it is not the answer anyone assumed:

It reads global tone. It is an exposure meter and a veto — excellent at killing a black frame or a fog whiteout, and constitutionally unable to tell you whether a photograph is well made. Every question about composition put to it comes back answered in brightness.

Thirty builds, shot from the identical camera in both lights, judged by a person rather than the model. Daylight wins, and the reasons are all about legibility: it rakes shadows across terrain, it puts white caps on the water, and it makes build materials readable — which for a gallery of what people built is most of the job.

Twenty-nine of thirty. And one exception, which is the photograph at the top of this page — a pier settlement whose lamps do more for it than the sun does.

Twilight loses on legibility and keeps its place anyway, for a reason that has nothing to do with the camera. It is the only light in which the builders' own lighting is visible.

Braziers along a wall. Candles on a table nobody eats at. Groundtorches marking a path between two buildings that a player already knows the way between. None of it is necessary — Valheim asks for light only to keep the dark from being dangerous, and these are far past that. Somebody placed them so a house would be pleasant to come home to after dark.

So the rule is not "shoot golden". It is golden leads, and twilight is the second frame on the builds that earn it — and no special case is needed to make that happen, because the model already sorts daylight first for its own unrelated reasons.

Which leaves one question: which builds earn it. The honest answer is that the pipeline cannot tell yet, and the reason is a good one.

A twilight frame is worth taking in proportion to how much light the builder placed. That is the first targeting rule this project has had that is a mechanism rather than a correlation — you can say why it is true before measuring it.

It cannot be computed, because the scan does not know most of the lights exist. The vocabulary of what counts as a light was written from Valheim's craftable build menu. This world is built from the prefab table, and the most-placed light sources in it are not craftable at all:

Around two hundred thousand light sources, invisible to the thing that decides what to photograph. The same blind spot had already been found once, in the furniture: three chairs that players place constantly were missing from the seat vocabulary for exactly the same reason, and the interior planner had been quietly walking past them for months.

• The light count. Until the fire vocabulary is repaired by hand — and it has to be by hand, because a pattern match for fire swallows a creature effect — twilight is a blanket re-shoot rather than a targeted one.
• The judge still cannot see composition. Everything in this article about framing rests on human verdicts. Pairwise comparison on a model with opinions is the unbuilt next instrument, same as it was at the end of part two.
• Blue hour proper. 0.71 is dusk, not the narrow window where the sky still holds colour and interior light reads strongest. The hour was chosen because interiors had already proved it, not because anyone searched for the best one.
• These are still other people's houses. Every frame stays published: false until a person promotes it. Nothing here consents on anyone's behalf, and the lighting is the most personal thing the pipeline has photographed so far.