We write down what we learned, as it was
A dated record of what we confirmed while setting the capture protocol, and why. Where our judgement was wrong, we correct it here rather than remove it. We think this is the record buyers actually read.
The aeffer-v1 schema — why pose_source and the /derived/ split exist
Mix measured and estimated values in one field and the buyer cannot verify quality. We separate provenance into its own field.
Capture rigs, three tiers compared — mocopi · HaritoraX · Rokoko · Xsens · Vision Pro
T2 is the minimum configuration that can produce wrist SE(3). T1 is for practising consent and rights, and falls short of sellable quality.
Why glasses alone cannot be sold — hand pose quality thresholds and monocular depth error
Monocular depth error of 5–11cm. Real-robot success drops from 95% on stereo to 45% on monocular.
Why DRM is impossible on training data, and what to do instead
Data absorbed into the weights cannot be recalled. We go with three things: compute-to-data, contracts, and a per-customer fingerprint.
Full write-ups will be published in sequence as we tidy them. If you need the detailed reasoning on a particular topic sooner, get in touch.