Expert work.
Captured with context.

The goal is not simply to record movement. It is to preserve the judgment behind the work: what an expert noticed, what they tried, and how they confirmed the result.

The proposed participation model starts with explicit consent, agreed recording boundaries and compensation for participating workers and businesses. Qualifying licensing arrangements may also provide participation in the value created.

Initial focus areas include heavy equipment, HVAC and marine mechanical work, beginning on Vancouver Island.

Development is informed by requirements discussions in the physical-AI data market. We are evaluating synchronized first-person capture, equipment context and outcome documentation, with collection and delivery requirements defined around each brief.

Field testing must establish whether those records are reliable, useful to model builders and economical to produce.

Commercial capture programs and research are establishing a market for learning from human experience. Mecka announced a $60 million Series B in October 2026, while Mercor publicly offers paid workplace-video collection for robotics training.

Nature Will reports pretraining Natura-1 on more than one million hours of egocentric human experience, combining video with reconstructed motion and task annotations. Across a training range of roughly 44,000 to one million hours, it reports a 67.7% reduction in robot-action prediction mean squared error on held-out robot data using teacher forcing. This is a prediction-error result—not a 67.7% increase in real-world task success.

The company also identifies complex activities and specialized settings as underrepresented in large-scale collection. FIELDKIT’s proposed focus on skilled diagnosis, corrections and verified outcomes addresses that opportunity; its usefulness still needs to be tested with buyers.

These company-reported developments support the broader opportunity. They do not validate FIELDKIT’s implementation or imply any affiliation or endorsement.