🔢 Floating-Point Precision Lab: Machine Epsilon, Number Spacing and Summation Error
Walk along a Blender-built number rail where representable binary floating-point values thin out by half every binade, round a real number to the nearest one, and watch naive versus Kahan summation drift.
⚙ Under the hood
A Blender-built number rail shows how binary floating-point values thin out at every power of two; round a real number, then compare naive and Kahan summation against the exact result and the u|x| error bound.
floating pointIEEE 754machine epsilonrounding errorKahan summationnumerical analysisBlenderThree.js
3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install