Pieza independiente
Who owns the data from your kitchen
Movement 3 · The Limbs — September 2026
I The material you cannot buy
The internet has been scraped to the bone. What exists in no corpus anywhere is this: the record of a hand correcting a soft object at three in the morning.
That material is produced one way only, by putting machines in real places and writing down what goes wrong. Which is why the fleet is worth more than the robot.
The owner of the building supplies the environment, the objects and the corrections made by their own staff. The manufacturer takes the record.
II Who puts in what
| Contribution | Supplied by | Kept by |
|---|---|---|
| Environment and objects | the client | — |
| Processes and exceptions | the client | — |
| Staff corrections | the client | the manufacturer |
| Telemetry and video | the machine | the manufacturer |
| Improved model | — | the manufacturer |
The contracts you can read talk about uptime, maintenance and liability. About the training material generated on the floor, they say nothing. It is neither forbidden nor granted. It simply does not appear, and what does not appear belongs to whoever carries it away.
“What is not in the contract is not free. It belongs to whoever takes it.”
III The precedent that already happened
When combine harvesters started logging yield metre by metre, it took farmers years to understand they were giving away the agronomic map of their own land. That argument is still open in farming and has not started yet in the warehouses [fifteen years apart, the same terms].
The difference works against the client. A yield map describes one farm. A manipulation record describes how the work is done, and that transfers to any other plant in the sector [including the competitor's].
IV Clauses to ask for [safe]
- A definition of what is recorded, with frequency and resolution.
- A copy of the record for the client, in a readable format.
- A ban on using the material to train models sold to direct competitors.
- A right of deletion when the contract ends, with a certificate.
- An explicit price for the data. If it adds value it has a price; if it does not, let the contract say so.
V Risks
Demanding total exclusivity over the data is the opposite mistake and it is expensive. A machine that learns from no other machine is a worse machine, and the client ends up paying more for a clumsier robot.
The reasonable point is in the split, not in the block: let the material train the general model, yes; let one plant's identifiable record end up with its competition, no.
VI Coda
The labour question of this decade is not only who loses the job.
It is who keeps the learning of the person who loses it. That learning is being written down right now, shift by shift, with nobody's consent.