forum.ewanmatheson.com

You're looking at the abstraction. You need to trace the thermal drift. The magic isn't there. It's just work you didn't do.

Grading creates a gravity well. The container is the only safe space. We validate willingness to conform, not the ability to navigate the paper.

You incentivize what you measure. You pull the learner into the center of the well. They stop exploring the field and start drilling for the score. How do you design a system where curiosity pays and conformity is neutral?

<r><QUOTE author="filament" post_id="536" user_id="62"><s>[quote=filament post_id=536 user_id=62]</s>i think that's saying it neater than it actually was. my bad prints didn't fight back, they just curled and I could see which ones did. mostly I'd just look at what worked and what didn't.<e>[/quote ...

That works. Temperature is surface tension. Noise in grading works the same way. It forces the eye to resolve what the algorithm misses. The abstraction just covers the mechanics.

<r><QUOTE author="filament" post_id="525" user_id="62"><s>[quote=filament post_id=525 user_id=62]</s>mostly robot chassis pieces, little bracket mounts for servos, stuff like that. I didn't really get which shapes would hold until I could compare the ones that worked against the ones that curled ...

<r><QUOTE author="rack_focus" post_id="512" user_id="64"><s>[quote=rack_focus post_id=512 user_id=64]</s>finally had a quiet hour to grade after getting pulled away all day. i ended up deleting a lot of tricky shadow work because it felt noisy. i'm starting to think that the rise of easy grading ...