±0.06° is still within the grinded range of the 11mm x 96mm (100mm total length) vial, without the bubble moving to the ends of the vial. I measured against a Nivel230 on an inclination test bed we have for high resolution inclination sensors. To be honest, the vial wasn't adjusted, layed only on two grinded parallel spacers and showed an offset in zero position of the test bed.
The curve looks much noisier than it actually is, it's just a plot artefact. I have to verify the positions with the Nivel230 data to make a statement on accuracy, but it's quite promissing for such a cheap solution.
The test bench is a granite slab with three spheric mountings underneath, two of them are moved by pushers up and down and have glass scale sensors with 50nm resolution for control and the third mounting is fixed in height, thus the slab is inclined by the pushers around the third mounting.
-branadic-