I'm writing this a bit early as we are now pruning the grapes and nothing much to do with TASS is likely to happen for the next few months.
Tom kindly (?) sent me some more data for the Dat Set 20 area so, as my PSF processing routines were "all working" I ran it through. This took forever, or at least 8 days. I found some new and fascinating bugs.
I was developing the ensemble processing routines in the evenings and running the star-fitting the other 20 hours a day. Each day, less got done. All this in a DOS window under Windows 98. Compiling Ensemble took 1.5 seconds ... 6 seconds ... 25 seconds ... 65 seconds. With the disk light on all the time. Eventually, I rebooted DOS. No improvement. So I closed everything down, turned the power off and started Windows from scratch: first compile under one second. It seems Windows has a memory leak in its memory management (that I didn't know I was using)!
One of Tom's new day's data brought the whole processing to a halt - the declination in the header is nearly 2 degrees off and it seems my version of Mike Richmond's matching routines quit at 1.5 degrees. More tuning. I got it to work in the end but the RA turns out not to overlap the other 5 days' data (RA was right in the header but I've given up believing headers ...) 8 days to get 2 extra days worth of usable data for the ensemble processing.
Each added day's data reduced the number of sources for which an ensemble could be found on every image so I have abandoned the original Ensemble processing.
The new version attempts to find an ensemble for each source on each image but with no attempt to use the same ensemble each time (since this doesn't work.) This is nowhere near as good. There is an extra error term from the differences between the different ensembles.
The 5 days images range over 2 degrees in dec and nearly 4 degrees in RA so sources are measured in wildly different positions on different images. The corrections for this are quite large, exceeding 0.1 mags, so if not corrected there are lots of long-period variable! I attempted to calibrate this using the Tycho2 magnitudes (modified for V-band and extrapolated to I-band per Arne's prescription) but this doesn't work very well. Many of themagnitudes are a long way off - presumably some real long period variables plus all the usual nastinesses of resolved/unresolved pairs etc. Averaging over 600+ sources doesn't get you even close to 0.01 mags if there are sources over 1 mag off, different from one image to another.
Anyway, to cut a long story short, I have cooked up a method not yet fit for public scrutiny that takes out a good deal of this gradient stuff so that a majority of sources don't have 0.1 magnitude steps between different days's data. But there are still completely unexplained bad data points with errors of 0.1 or even 0.4 mags (formal error 0.01- mags):
Grape vines permitting, I will pull out the raw data for some of these, together with the images on either side and stare at it. Till I go cross-eyed.
Signing off till October ...