Unresolvable at Hallow Gate

The array is not a detector. It is worth saying that first because this file is otherwise easy to misread. What Instrument 04 does is measure how much of a picture it cannot make sense of — how dark a region is, how much it changes between frames relative to its own brightness, and how little structure it holds. That figure is the resolution floor, and it is a measurement of the camera and the room, not of anything in the room.

A well-lit interior gives you a floor in the teens. An unlit cellar gives you ninety. The whole point of the number is that it is boring and predictable and you can move it by turning a light on.

At Hallow Gate you cannot move it by turning a light on.

The building

A four-storey brick warehouse, empty since 1997, sound and dry. The Survey was there for three nights at the request of a developer's surveyor who had lost two contractors off a job and wanted a report he could put in a file. We measured eleven positions and found a building with old wiring and nothing else in it.

The service stairwell runs the full height at the north-east corner. It is concrete, it is painted a pale institutional green, and it is lit by original fluorescent fittings which work. Standing in it, it is an ordinary, slightly grim, entirely visible stairwell. You can read a newspaper in it.

Where the figure comes from

The floor is the peak of three terms and it is worth breaking this one down, because the breakdown is the whole of the finding.

Darkness contributes almost nothing in the stairwell. At 340 lux the tiles are well above the level at which darkness dominates the score, and on the second night the darkness term was lower than in the main hall.

Flatness contributes something. Painted concrete is flat, and a stairwell wall in a single colour genuinely does hold less structure than a warehouse floor. But flatness is measured between neighbouring tiles rather than inside them, and a stairwell has handrails, treads, nosings and a change of plane every metre. It is not featureless and the term does not read it as featureless.

Grain is where the figure comes from. Grain is the frame-to-frame change in a tile measured against that tile's own signal — sensor noise, essentially, expressed as a proportion. It is high in the dark because a dark tile has almost no signal to divide by, and it should fall away sharply as you light a scene, because the denominator grows.

In the Hallow Gate stairwell, at 340 lux, the grain term sat at a value that a sensor at that exposure should not be able to produce. We put a lamp on it and the picture got brighter and the noise got brighter with it, in exact proportion, which is not what a camera does.

Afterwards

Both array heads went to the bench on 14 December and were run against the standard target under controlled lighting at four exposures. Both were within specification on all four. The 300 W lamps were checked for flicker at the array's frame rate, which is the one failure mode that could plausibly produce this, and found to be running from a smoothed supply with no modulation above 0.4 per cent.

The building was demolished in 2016. The Survey did not get back into the stairwell with a fourth-revision head, which is the equipment we would want for it now, and that is the Survey's own failure rather than anybody else's.

What we can say

We have a stairwell in which a camera behaves as though it were in the dark, while being in the light, on two heads, on three nights, with the reference photographs in the box showing an ordinary lit stairwell.

We have no reconstructions to report. The array did put figures into the region — that is what it does above the gate, it is drawing and not detecting, and the count is in the file because we count them separately and never add them to anything. They are not the finding. The finding is that the gate was open at all in a room you could read a newspaper in.