ASIAA reported during the Sept Engineering Run post-mortem telecon 2021-10-04 that some fiducial fibers were not correctly identified. This may have affected the cobra convergence performance.
If the geometry is close it's not hard to get all of the fiducials right, so is this a bootstrap problem? Either way, we need a command to find the fiducials and update the pfi <--> mcs mapping; there's code in the `PfiTransform` object to do this, but if Jennifer has a better implementation we should definitely do this.
Note that we should find the fiducials without the back-illuminator until the time that the ab-initio guess on the transform is good enough that we can use a matching radius small compared to the distance from the fiducials to the nearest cobra. Or add some smarts to iterate, but simply turning off the back illuminator is easier.
Either way, operationally we need to be able to slew to a field, identify the fiducials, update the mapping, and generate a valid PfiTransform object. I think that the procedure should be
slew to field
illuminate only fiducials
create/update PfiTransform object
turn off fiducials for the rest of the visit
turn on back illumination
converge cobras to the correct places
turn off back illumination
[and, while we're learning about the PFI:
illuminate fiducials
take an MCS exposure
turn off fiducials
so we can learn about stability. We may not actually need to so this]
It should be true that this object is then valid for the entire observation; if the exposure time is long enough and the telescope is tracking this may not be true.