Big personal milestone: my first arXiv paper is live. bbsolver began as a production problem I kept hitting in motion design: dense sampled animation is safe, but awful to edit. Every frame becomes a key. Handles get noisy. Animated vector paths become painful to clean up - as a result. I built the tool I always wanted. bbsolver converts dense sampled animation into sparse, editable keyframes under an explicit error budget. The key idea is treating timing and shape together: it chooses key timing and, for vector paths, a topology-consistent path representation under one validation budget instead of doing temporal key reduction and spatial path simplification as separate passes. Results: • After Effects DUIK walk cycle: 12,684 samples → 540 keys at ε=1, a 23.5× reduction, with AE playback under 1 px / 1° max error • Ant rig: 11,956 samples → 653 keys, 18.3× reduction • Blender-sampled FBX mocap: 13,455 samples → 214 keys at ε=3, a 62.9× reduction • Matched-accuracy baselines needed 4.5×–27.5× more scalar key entries • AE-compatible procedural path compression reached 6.7× The paper is on arXiv under cs.GR, and the solver + reproducibility artifact are public. I’m especially proud of this because it came from actual production frustration, not from trying to invent a paper topic. It’s a motion-design tool that grew into a graphics systems paper. Paper: https://arxiv.org/abs/2606.09741 Repo: https://github.com/ivg-design/bbsolver