WHAT WE CAPTURE
Consider a single, ordinary moment of skilled work: an electrician torquing the neutral lug in a residential panel. It takes perhaps eight seconds. In those eight seconds there is more useful information than in an hour of the staged household video that today's robot models are fed, if you know how to capture it.
We record that moment four ways at once, on one clock. A head-mounted camera sees exactly what the electrician sees. A microphone hears him explain, in his own words, what he is doing and why (“setting the driver to forty-five inch-pounds, per the panel schedule”). The instrumented screwdriver reports the actual torque curve as the lug seats, peaking at 45.2 inch-pounds, squarely inside the specification band. And every one of those signals lands, timestamped, in a single training-ready record.
The result is what we call behavior-grade data: not sparse clips of finished work, but a continuous, synchronized account of how skilled hands actually do the job.
WHY FORCE MATTERS
Most of the video data available to robotics developers today has no force in it. A camera can show a hand turning a screwdriver; it cannot show how hard. The published work on contact-rich manipulation points consistently in one direction: adding a genuine tactile signal to vision measurably improves what a policy learns. Force is the premium signal, and almost nobody is capturing it.
Compare that to the gig-sourced video collections on the market today, which capture household chores by the hundred thousand hours (with no force signal, no expert narration, and no trade skill in the hands being filmed). That data is abundant precisely because it is easy. Ours is valuable precisely because it is not.
A CLEAN CHAIN OF TITLE
Just as important as the data is where it comes from. Every hour we capture is collected under a signed consent and royalty agreement with the individual electrician, so the professionals whose skill trains these models are compensated participants, not unwitting subjects. Their licenses are verified; their identities are protected; their royalties continue as the data earns.
For the laboratories that license our datasets, this is not a courtesy; it is the chain of title their counsel will insist upon. We built it in from the first hour of capture, because provenance cannot be added to data after the fact.
FOR CONTRACTORS
The same capture that produces training data produces something an electrical contractor can use the very same day: photographic and video job documentation, a task log, and a record of the actual torque applied at every termination. NEC 110.14(D) requires that connections be made with a calibrated torque tool; where an owner or a commissioning specification also calls for evidence of the values achieved, that evidence has already been captured. Our contractor partners receive these records for every captured job.
GET IN TOUCH
If you are building robot foundation models and would like to see a sample of our capture, or if you are an electrical contractor who would like to have your electrician work automatically documented, or a licensed electrician curious about being paid for your expertise, we would be pleased to hear from you.