CDO

I work on an ensemble grib of the canadian GEPS model.
I've united the different members into one grib (see the 200KB grib attached).

when running wgrib2 on it, I see the following 21 lines corresponding to 21 enesemble members:

1:0:d=2023092400:TMP:2 m above ground:0-0 hour ave(99:99 hour fcst):ENS=? table4.6=255 pert=0
2:9656:d=2023092400:TMP:2 m above ground:0-0 hour ave(99:99 hour fcst):MM-ENS=1
3:19312:d=2023092400:TMP:2 m above ground:0-0 hour ave(99:99 hour fcst):MM-ENS=2
4:28968:d=2023092400:TMP:2 m above ground:0-0 hour ave(99:99 hour fcst):MM-ENS=3
5:38624:d=2023092400:TMP:2 m above ground:0-0 hour ave(99:99 hour fcst):MM-ENS=4
...
...
19:173808:d=2023092400:TMP:2 m above ground:0-0 hour ave(99:99 hour fcst):MM-ENS=18
20:183464:d=2023092400:TMP:2 m above ground:0-0 hour ave(99:99 hour fcst):MM-ENS=19
21:193120:d=2023092400:TMP:2 m above ground:0-0 hour ave(99:99 hour fcst):MM-ENS=20

Then I try running "cdo timmean <grib_path> <output_path>" I get that the output is equal to the input, as if cdo 1.9.9 doesn't observe that all these 21 lines belong to the same timestamp and field.
While on my other server I have cdo 1.9.3 which manages to created the ensemble mean perfectly! (SEE the "CORRECT" grib attached).

See CDO version details (1.9.3 vs 1.9.9rc2) in the attached files).

I hope that you can help me understand what has changed between the versions that's causing the same command to fail. and advise for an alternative that works on cdo 1.9.9.

THANK YOU!