I confirm the flux calibration is indeed bad. I picked visit=111775 from the GE visits and the flux calibration vector looks awful. This seems to come from bad pfsMerged in spectrograph=2. Robert pointed out that n2 has inconsistent fluxes between real data and targetType. I confirm that it is only n2 and the other arms+spectrographs are OK. This seems to be a detectorMap problem.
-----------------------------------

#fiberId targetType b2    r2      n2
658 FLUXSTD 2572.8 5661.3 1606.0
659 FLUXSTD 1805.8 4226.6 405.5
660 SKY 53.7 116.2 494.0
661 SKY 52.1 116.8 1791.0
662 FLUXSTD 1539.3 3674.8 459.9
663 SKY 45.2 112.0 451.9
664 SKY 53.6 124.6 429.4
665 SKY 44.2 114.1 1393.1
666 FLUXSTD 1227.6 2884.7 529.0
667 SKY 39.7 105.4 499.6

-----------------------------------

The latter 3 columns are the median flux in each arm and is from pfsArm.

Although a complete check is not yet, here is an example of flux calibrated spectrum (a FLUXSTD) before and after the fix. I looks the calibration is significantly improved.

Left: before the fix, Right: after the fix

Multiple visits for the same object.

Masayuki Tanaka Are we going to keep investigating the flux loss issue (of about 20-30%) in this ticket? or would PIPE2D-1580 be a better place?

This ticket is about the bad fluxCal due to off-by-one problem in the detectorMap. So, we can safely close this ticket. For the 20-40% flux loss issue, I will file a separate ticket. There is also the nearly-no-flux problem but we can track that work on PIPE2D-1580.