Thanks to 绿茶字幕组 for allowing us to mux in their carefully crafted 中文 subtitles.

This is a high effort filtering and encoding for Champignon no Majo.
slow.pics compare: https://slow.pics/c/15vrEyFa

The BD is very nice. There are very few cases where the WEB contain more information than BD. We perform a MinMax merge of WEB and BD on luma in the episode. Essentially, if the BD has more information than WEB, we do nothing. If the WEB has more information than BD, we allow WED in. The WEB is decheckerboarded before joining with merge.
In the OP, as usual, we perform a Lehmer merge of multiple episodes of BD. The OP merge is nice, but contains quite heavy blocking. It also demands way more bitrate than the main episode, so instead we perform an additional light vsdenoise.deblock_qed on the OP to reduce the blocking and control the filesize.

The main episode is descalable. We descale the main episode using vsscale.Rescale, and then we write a custom akarin.Expr to carefully dampen the sharpness of the descaled image. Thanks to many people in the JET server such moe teaching me the method, we fully recover the layer of 1080p texture on top of the image after descale.
The OP and ED of this anime is not descalable. Instead we use a stronger vsTAAmbk.TAAmbk to deal with the OP, protected by a custom akarin.Expr taper. And we use vsaa.based_aa to perform aa on the ED.

This anime has a static layer of texture on top of the image, which is really hard to encode. This layer of static texture also makes it impossible to denoise protectively due to this layer of texture having different motion vectors than the underlying content.
Without denoise, we resort to tuning the encoder to try to control the filesize at or around 2.0 GiB.
Specifically to control filesize in a source that's very static as this, we emphasise inheritance by boosting P frames and dampening B frames. Based on vsmuxtools.settings_builder_x265, we set bframe_bias=75, ipratio=1.3, and pbratio=1.4 for the encode.
To alleviate blocking in the case of intra mode being used on noise in B frames, where there isn't enough quantisation to support encoding the noise from scratch, we slightly increase psy_rdoq to 3.0. And to overall take care of blocking both from the source and from encode, we set deblock=[0,0].
At last, we add a basic layer of dithering noise via aom_film_grain (not included in the slow.pics compare above).

Thanks to all the people creating the tools, plugins and vsfunc's that make this possible, and thanks to everyone in the community for teaching me, pointing out issues, and giving me advice.

You are welcome to do anything with this encode, including remuxing and republishing it. However, we would ask you to leave a credit to Kekkan if you can.

We recommend using MPV to playback our releases. Thank you for seeding after download.