The following plots show the median positional error in pixels (`dx`) for each fiber, meaning each point in the plot represents the median value for a given fiber for a given visit.

The plots are split up so each page shows a single spectrograph for a given run (run08, run11, run12), each row shows the arm, and each column shows the calibration version (v3, v5, v6). The points are color-coded according to their label (e.g. "sky", "calspec", "ga", "ge", etc.).

Most of these look fairly "normal" (i.e. median residuals near zero) except for the single attached plot ("Figure 13"), which shows spectrograph=3 and run12 and shows that we did not really do well for the sky fields during run12 in any arm, compared with the full pdf which shows we did do alright in the sky for the other runs and spectrograph.

"Figure 15" is a copy of this plot but split up by the "status" (e.g. DETECTORMAP_USED and DETECTORMAP_RESERVED).

fiber-median-by-spec-dx_median-label.pdf

What is the unit of dx_median? Pixel?

> What is the unit of dx_median? Pixel?

Yes, sorry, we agreed to put things always in pixels. I have it listed on the top of the full PDF I think but I'll put it on the axis labels as well, thanks.

The dominant feature in Figure 13 above was found to be the two out of focus visits.  The following plot is the same thing but with those visits removed (97484 and 97485).

However we also started to question why the out of focus data would cause the data to behave in a non-symmetrical way (see plot below) and why the n band looks like it was processed as two separate halves (it shouldn't be, according to price). Currently arnaud.lefur is looking into this issue.

Figure 16 shows the data with the out of focus visits removed, which then also reveals a feature in the center of the n band. This seems to correspond to the faint mtp bundle so we are not investigating further. The following plots show the same feature present in the spectral (i.e. dy in pixels) offsets:

Zoom:

Yes so I was able to look at those visits.
A bit of context first, not only the slit was defocused but also rotated along y-axis (x-axis being focus, z axis vertical)
meaning that one side of the slit was closer to focus and the other side further away.
In the spatial direction it means that the slit has a bigger apparent size.

You can see that the right part of the curve is close to 0, then at you go left, the spots are more and more defocused, and moving to the left.
At some point the detectorMap is bad enough, that the fiber identification is wrong, spots are then assigned to the neighbouring fiber, which is explained the sudden discontinuity.

The DetectorMap QA should have detected that the spot IDs are off, meaning that some spots weren't found. So that's another thing to track.

We saw this before when a fibre was mis-identified near the left edge of on of the right-hand CCDs, leading to odd distortions

Attached are the median dx offsets in pixels for every 20th fiber across all visits in EDR2.

Ideally the offsets should all be noise around zero, ideally improving with each run, which doesn't appear to be the case for most of the detectors.