The workflow was extract frames into pictures, delete all smudgy interpolated frames (even # frames in this case), remux back to a video with editing codec, interpolate to higher fps than final goal (i used 120), at some point extract the audio from original and add back with delay if needed, compress into final package while dropping any extra fps
Further reading: I use Apple ProRes codec set to 4444 and force colorspace label to Rec. 709, then interpolate with Flowframes output format set to "MOV" "ProRes" "4444" "YUV 4:4:4 10-bit." Doing all of that avoids the colorshift that seems to happen with FFMPEG programs that accidentally convert colorspace from Rec.601 to Rec.709