Due diligence: where the data comes from, and what it never contains.

One row for each robot-training role (teleoperators, robot data collectors, demonstration and first-person video recorders, robot data annotators, autonomous vehicle and drone test operators, robot fleet technicians) that an employer posted on its own public job board: the employer's careers site or the public job-board feed its applicant tracking system publishes (Greenhouse, Lever, Ashby, Workday, Amazon Jobs and others). We read every board every day and keep every role we have seen: 1,197 roles at 107 employers since 12 Sep 2026.

A role is in the role files only when we read it from a job board that anyone can open with no account and no automated check to get past. An employer we can read only some other way stays on our public job board and in the market's daily totals as they were recorded, and never appears role by role. No account, no paywall and no automated check is ever passed to read a board. Today that leaves Tesla out of the role files: we read it another way for our job board, so it appears only in the market's daily totals.

No. The files describe jobs, never people: no posting text, no names, no email addresses or phone numbers, nothing about the people who apply and nothing about the visitors of our own site. A role's fields are its employer, title, kind of work, posted pay, place, the countries it is open to, its contract type and its dates.

No. Every field is what the employer itself published on a public job board at the time we read it. Nothing comes from employees, insiders, private channels or paid access, and nothing is inferred from a source that is not public.

Role records begin on 12 Sep 2026, when our reading began; nothing before it is backfilled. Each role keeps the day we first recorded it and the day we last saw it, as recorded on those days, and roles that came down stay in the files (172 of them today), so the record has no survivorship gap. Every close, reopen, pay change and title change is logged with its time from 29 Oct 2026; before that a role shows its latest state. A role an employer already had up in its first week in our data is marked as its back catalogue, never counted as newly opened.

Every employer is read the same way and every role is classified into the same kinds of robot work. The market's daily file has 48 days with no gaps. Each day we rebuild the open-role count from the role files and compare it with the count recorded that day: the median gap is 0.2%. Employers are added as we find robot-training boards; each employer's own days show when it entered.

Six files, as CSV or as JSON: the roles, the change log, each employer by day, the market by day, the employers (with the stock listing of a listed parent), and a dictionary with every column's meaning and how many rows carry it. They are rebuilt every day at addresses that are the subscriber's for the year, each stating when it was built. The data points and their counts are on the radar's page.

For the one firm that subscribes, to use inside that firm for twelve months: never resold, passed on or published beyond it. The licence is written out in full. $25,000 a year for one subscribing firm.

Yes. The sample is October 2026's files, taken as a snapshot on the day you ask, sent to your work email from the radar's page. A mutual NDA can be signed before it is sent.

RobotTraining.jobs, a job board for robot-training work, operated by O-mega Enterprise, Inc. The radar launched on 29 October 2026 and is built from the same daily reads as the board.

Each subscriber reaches its files at addresses carrying a key signed for that firm, valid for its licence year; downloads are limited per key and per connection, and no third party receives the files.