Best practice for cyclic WRF-3DVAR: using wrfinput or wrfout as first guess?

[paparao]

Dear WRFDA community,
I am currently setting up a 6-hourly cyclic WRF-3DVAR data assimilation experiment and would appreciate guidance on best practices for operational cycling.
My workflow plan is:

• Run WRF real.exe to generate initial wrfinput_d01
• Use this as first guess (FG) for the first 3DVAR cycle (00 UTC)
• Run da_wrfvar.exe and generate wrfvar_output
• Use DA output to run WRF forecast up to next cycle (06 UTC)
• Continue cycling at 00–06–12–18 UTC

My question is regarding the first guess for subsequent cycles:
For the 06 UTC and later cycles, which is recommended as FG for WRF-3DVAR?

1. Use wrfinput_d01 updated from previous DA output
2. Use wrfout_d01_YYYY-MM-DD_HH:00:00 from short WRF forecast as FG
3. Any other recommended operational approach?

I would like to follow the scientifically correct and commonly used method (NCAR/NCEP/operational centers) for cyclic 3DVAR assimilation.
Any guidance or example workflow for best practice cyclic WRF-3DVAR implementation would be very helpful.