Work down this list in order. It is arranged by how much of the roll the cause explains — start at the top, because a wrong denominator makes everything below it look like a different problem.
#1. Is there an empty gate for this session?
If not, stop here. Without it the density has no scale and every frame is an estimate. Shoot three gate exposures under the same lighting and re-run before touching anything else.
#2. Is the rebate in frame?
The film base is read off the clear film around the image, once per roll. Cropped in camera, there is nothing to read. This shows up as a roll that is consistently off rather than randomly off.
#3. Is the light the one that was measured?
Only one light source has a measured calibration in the app: scanlight-v4, at 665 / 525 / 450 nm. If yours is different — even "the same LEDs, different batch" — the numbers are being read against somebody else's calibration, and the app will not tell you that on its own. Post the rig in Rigs.
#4. Did the three bands of a frame move relative to each other?
The three separations are resampled onto each other rather than nudged to the nearest pixel, and every roll reports how far each channel had drifted. A large drift on one frame is usually the stand, not the software. Look at the reported drift before you look at the colour.
#5. Is it the whole roll, or some frames?
This is the diagnostic question, and the two answers point in opposite directions:
- Whole roll, same direction — a denominator problem. Causes 1 to 3.
- Some frames, not others, same settings — that is the one this software exists to remove. If it still happens after 1 to 3, I want to see it.
#6. Is it the emulsion batch?
Film base is a property of the stock's orange mask and its development, so a table can only ever approximate a particular roll — two rolls of one stock have measured up to 0.033 D apart on this track.
There is a measured batch correction with a slope of [1, 1.8, 1] across R/G/B, and the green figure was confirmed independently on two stocks — ColorPlus and Vision 250D. So a roll that is off in green specifically, and consistently, has a real suspect rather than an excuse.
#What to post here
Bring measurements, not adjectives. Useful:
- what light and camera,
- whether the gate was shot for that session,
- the correction the app landed on, which you can paste as a block:
That renders as the same axes the app shows, so we are looking at the same instrument rather than describing it to each other.