If you have time for this that would be great, thank you.

This ticket got expanded slightly so that the code that makes the plot could be reused to make other plots of the same format.

The four line histograms on the right come from the points deliminated by the quartiles marked with the arrows of the corresponding colour.

Thanks for updating stuff, it prompted me to finally take a look at this!

I left some code-related comments on GitHub for you, don't hesitate to ask if anything is unclear.

From a plot perspective... can haz captions? There is a lot happening in each of the plots you shared and I honestly have no idea what the main takeaway from each plot is supposed to be without some explanation. That may mean the plots are too complicated... and/or it may just mean you need to give me a little more context about what the fakes are that you're inserting and what you're hoping to learn from the plots.

General Plot Information:

This code makes plots of a given array of x against a given array of y with the y distribution collapsed into a histogram on the right hand side of the plot. The examples on this ticket are fake magnitude against input fake magnitude - processed fake magnitude, log 10 ( blendedness)  against the same mag difference and distance to nearest neighbour against magnitude difference. For each of the plots the points are given as either blue circles (included in the median and sigma MAD quoted) or blue crosses (not included in the quoted values), the regions of the plot within 1 sigma MAD and of a high enough density are shown as a 2D histogram to allow structure to be seen more clearly. The median and sigma lines on the plot are from calculating the median and sigma MAD used in each vertical bin. Half, one and two sigma MAD lines are shown in lieu of contours to illustrate the distribution. The vertical lines show the quartiles and the arrows indicate the colour of the associated line histograms on the right. All the plots include only objects input as stars.

Information on the individual plots:
All the plots use fake sources inserted during the processing. Fakes can be inserted either into the coadds or at the visit level which was done here. The input magnitudes are the magnitudes that were given to the insertion code in the catalogue of fakes and the output magnitudes are those recovered during the processing.

Magnitude vs Magnitude Difference

This shows how the magnitude difference (between input and recovered) changes with input magnitude, unsurprisingly it gets larger for fainter sources. There is a small offset, I do not know if this is important. The long term plan is to track it and see if it changes.

Blendedness vs Magnitude Difference

Similar to above but with log10(blendedness) on the x axis, again mostly unsurprising more blended things are less well recovered. Possibly a hint that really close pairs are more likely to have their magnitudes over estimated.

Distance to Nearest Neighbour vs Magnitude Difference

The distance to the nearest neighbour is calculated by matching back to the catalogue and taking the second nearest match, based on the assumption that the nearest match is the object itself, this matched object could be anything not just the stars used in the plot. This is plotted against magnitude difference again and also shows nothing particularly surprising.

SPs include those moved from DM-20477. The plot ended up on this branch.

Thanks for working with me on all the little back and forth things! I understand "fix pipe_analysis" is not something we can fully solve on this ticket, and I think the new fakes plots will be instructive and I really appreciate you doing a bit of useful cleanup along the way. I don't think you've touched anything here Jenkins cares about (cough needs some tests cough also out of scope cough), but please do a run just in case I'm wrong, and then feel free to merge.