The overall goal of the VLA QA criteria is to create a 'fast' vs. 'slow' lane dichotomy. In this scenario, fast could go through to accept and archive without needed to look at specific stages, while slow lane would go through the standard, in-depth weblog review that has typically accompianied the 'SRDP treatment' of VLA EBs. Precise QA scoring is not necessarily required, but could be helpful, in general anything that triggers slow lane should make the score drop below 0.5.
- hifv_importdata
- Include a plot of the API data during the observation
- if API or wind missing or out of spec, trigger slow lane
- Check for missing scans
- always look at
- If flux.csv is used, always triggers slow lane
- Include a plot of the API data during the observation
- hifv_hanning
- simple QA, it either worked or didn't; automatic score of 0 if hanning smooth didn't work for some reason
- hifv_flagdata
- if a flag template use, always triggers low land
- look for endtime = 0 in flag.xml (can pipeline do this?)
- look for '1858/' in online flags - trigger slow lane
- clipping > 1% with spectral line windows
- clipping > 5% with only continuum windows
- total flagging >30%
- hifv_setjy
- QA score of zero if no flux calibrator intent or non-standard flux calibrator
- hifv_priorcals
- add API information to the weather plot here
- weather station not functioning
- missing switched power data
- TEC maps presence can be safely ignore
- hifv_syspower
- >50% of all dataoutside the 0.7-1.2 range per band
- needs per Baseband selection syntax for mixed 8 and 3 bit setups
- hifv_testBPdcals
- >4 antennas have a DTS issue
- indicates bad reference antenna
- median delay per BB > 15 ns
- >4 antennas have a DTS issue
- hifv_checkflag (both for bpdcal and allcals)
- > 75% of data flagged
- hifv_semifinalBPdcals
- remove warnings for single antenna and single spw being completely flagged
- if delays > 15 ns per baseband give a warning
- some number of spws on some number of antennas newly flagged gets a warning
- hifv_solint
- short solint > long solint
- calculated median calibrator scan time, if long solint > 1.5x the median calibrator scan time provide warning
- fix gain calibrator long solint calculation (refine what's already in the pipeline using medians rather than means)
- hifv_fluxboot
- >50% of spws are fully flagged in the flux gain table
- >50% of all data flagged
- spindex > +/- 5
- flux scale fails on 1 calibrator out of several
- look for unity in CGC model column (indicator of flux density not being measured
- hifv_finalcals
- suppress the per spw flagging warnings
- delays>15ns
- very little TOS on BP calibrator left in gains vs. time for BP cal
- very low amps in some BBs/spws?
- BB for antenna has high phase spread (indicative of DTS that got through)
- phase scatter of short solint vs. long solint to determine coherence
- warn if >2 antennas have < 0.6 coherence on average
- look for large > 50 degree phase jumps between scans on the same complex gain calibrator
- would be nice to tie this in to API information
- high or low gains for whole polarization or >=2/3 spws in a single BB per polarization
- hifv_applycals
- have a cumulative table
- have a table per band
- in each table break out by CGC and target(s)
- >85% of data flagged
- >30% additional data flagged per target or CGC
- jumping of flagging of target > CGC
- hifv_checkflag(target)
- huge amount of additional flagging
- >10% antennas have excessive flagging deltas
- lose whole BB or high res. spectral windows completely
- hifv_statwt
- gigantic weights
- mean > 1,000,000
- when variance goes to scientific notation in weblog(explicit OOM? 10^4?)
- high weights
- individual weights > 150
- high variance
- variances > 1000*mean
- variance in scientific notation (need to know when)
- high increase in flagging
- current scoring may be based on flagging and need ot understand what is happening currently
- heuristic development to flag outliers would be desirable
- per-field/per-spw weight characterizations may be necessary too
- gigantic weights
- hifv_plotsummary
- devidations >10% of mean in amp vs. freq
- for > X% of the data ~10%?
- per spw, per antenna
- phase spread vs. freq - implies resolved calibrator
- plot of real vs. imaginary instead?
- round=ok, point-like
- structured = resolved
- plot of real vs. imaginary instead?
- devidations >10% of mean in amp vs. freq