VMAF experimental not up to the mark?

[ls-milkyway]

Used Batman 2022 trailer 5 for encoding using qencoder with target vmaf settings enabled:

min crf 15, max crf 60, test-steps 5, target 97.00 & model vmaf_float_v0.6.1.pkl

Other settings:
--threads=2 --kf-max-dist=240 --tile-columns=1 --tile-rows=0 --cpu-used=3 --good --end-usage=q --cq-level=23 --bit-depth=10 --i420
Split-mode= pyscenedetect.

After the encoding used ffmpeg with libvmaf: ffmpeg -i output.mkv -i ready.mkv -filter_complex "[0:v]scale=3840:1608:flags=bicubic[main];[main][1:v]libvmaf=F\\:/Audio Video/ffmpeg-all/Models/vmaf_float_v0.6.1.json:log_path=VMAF_out.json:log_fmt=json" -f null - to calculate vmaf scores mean vmaf score was 90.00??

Please note used bicubic to scale the encoded video for vmaf calculations as recommended by netflix.

1. Why am I getting mean vmaf 90 and very fluctuating graph: