USE CASE RQM

All notes are related to PR/REP only. It does not cover skims/SP/AWG

Before QA
=========
1)  data taken --> entry into IR2 tables 
2)  data processed --> entry into PR tables 
3/4) sweep --> tdf file produced
     Each subdetector checks status of each run
     and marks the quality

QA process 
==========
1) Process TDF file produced by sweep 
2) Consult PR tables to check if runs from tdf mark as "done"
    - may involve joining tables from SLAC/Padova 
4) Consult PR and IR2 tables to compare
   # events processed by PR and # events reported by IR2
    - would like to be able to get ratio (per stream/skim):
      processed events/total # events 
5) Visually look at QA plots. Catch problems (ideally all
   should have been caught by subdetector people).
   If something wrong, interact with subdetector people. 
6) Read AllEvents to verify e.g. if no obvious problems
   with storage 
7) Produce a list (ASCII format) of "bad runs".
   This list does _not_ include runs marked as
   bad in PR or by subdetector people. 
8) Produce final result: table(http format) used by 
   people doing analysis. Maps runs/blocks of data
   to tcl files containing collection names.

To summarize, involved tables/metadata:
 - TDF
 - IR2 tables
 - PR tables from SLAC
 - PR tables from Padova
 - QA plots
 - ASCII list of bad runs
 - web page table

Some further comments:

# events kept in:
 - TDF (# events per stream/skim)
 - PR table (# processed events, only per run)
 - IR2 table (# recorded events, after L3 trigger, only per run)

A collection/run can be "bad" because of many reasons
 - problem with sweep
 - did not pass subdetector QA
 - problem in PR
 - quality of reconstruction (tools to verify non-existent yet)
 - problem with storage (e.g. corrupted file) The new bookkeeping should
keep the status why collection/run is "bad".

In some cases, e.g. if storage problem, might be useful to keep some error
messages, or how to reproduce the problem.

If found file-related problem with one collection, should be able to
locate all collections in the file and check them.

What if AWG finds a problem? Should they report it to QA, then QA should
re-check, or AWG should be able to set status to "bad"?

SkimData can provide info similar to what is kept in the final table
produced by RQM. SkimData info not QAed, but has all collections,
including skims. The RQM table QAed, but contains only AllEvents
collections.

It was clear during the conversation with Eugenio that we did not fully
cover the topic. We are going to continue, so stay tuned.
==============================================================================