| | 8 | Eddie and I examined difference between his ubercal magnitudes and my dvo magnitudes after applying the ubercal zero points. We tracked down 3 main errors: |
| | 9 | # a bug in the -64bit image read byteswap code in dvo (used only for the ubercal table) messed up mid-level bytes. fixed this in dvo |
| | 10 | # a difference of how to treat off-chip positions for flat-field corrections (dvo saturated position on the chip, ubercal chose a random (incorrect) location). This only happened 10^-7^ times, but ubercal will adopt the dvo solution |
| | 11 | # ubercal uses per-exposure airmasses, while dvo uses per-detection airmasses. For the test field (near 0 Dec), the difference was quite small (< 1mmag), but the difference is important at higher airmass. We compared by applying the airmass difference, but in a re-run ubercal will move to per-detection airmass values. |
| | 12 | In the end, with the fixes and corrections, all detections agreed at the floating point limit (5e-7). |
| | 13 | |
| | 14 | I am now working on a (largish) test sample to run all steps to generate the reference database. This includes merging the synthetic data into the dvo database, and running relphot on everything. I will use the synthetic data to test the amount of clouds as an additional check. After the reference database is done, I will also apply astrometry to the resulting database to generate a dvo db for distribution |
| | 15 | |
| | 16 | I also ran another set of test MD image for Doug F. to further check on the faint end bias. |
| | 17 | |
