Physical Capture v1 · pay shown first
Film the chore. License the judgment.
JumpTask proved people will put on a head strap for $5–$20/hr. EgoVista proved robotics teams buy egocentric + pose/contact/action in LeRobot / RLDS. Primacy sits on top: capture like JumpTask, package the layers EgoVista ships, then bind objects, constraints, exceptions, and force into the Primacy Graph — royalties on reuse, not a one-off credit mill.
Casual lane
Clip missions
6–12 minutes. $1 first payout. Gift card or USDC. Pay only accepted hours. Rejects unpaid and dropped from inventory.
Sovereign lane
Physical Judgment + bundles
3 accepted clips bootstrap a Room. List as Physical Judgment / Egocentric Traces / Training Data Bundles. 85% USDC on Arc.
Missions — pay before you start
Prep a meal
Head-mounted. Cook or prep something you were already making. Narrate intent. Abort if a blade or heat is unsafe.
Tidy a surface
Clear a counter, desk, or bench you actually use. Hands in frame. No faces in mirrors.
Use a hand tool
Screwdriver, wrench, knife, or similar. Force judgment and an abort rule are the product.
Pack a fragile item
Box something that can break. Path, padding, and abort if it shifts.
POV capture protocol
Same rules EgoVista and JumpTask POV tasks use. Primacy adds abort judgment, instant usable-hour QA, and a graph bind.
Hands-free POV
Head or chest mount. Phone not in-hand. Start-to-finish at normal speed.
No faces
No faces, mirrors, plates, or screens with personal data. Blur before any video leaves the device.
Think-aloud track
Narration is a separate field. Licensed for model training only — no voice cloning or synthesis.
No MP4 here
This vault stores the episode card (LeRobot-shaped JSON). Pixel capture stays on your device until a license job completes.
Pay is shown before you start. Rejects are unpaid and deleted. Open Physical Capture · Footage licence
Usable-hour QA — published checklist
Instant gate on submit. Pay only if every check passes. Failures are unpaid and removed from the license inventory.
- Hands-free head or chest mount — phone not in-hand
- No faces, mirrors, plates, or personal screens (or blur-before-vendor)
- Hands in frame for the manipulation
- Start-to-finish task, 3+ steps, normal speed
- Duration inside the mission window
- No names, addresses, account numbers, or license plates
- Think-aloud track present (training licence only — no cloning)
- Abort / stop rule written
- Contact / grasp flags filled
- 2D hand description filled
Annotation stack
Robotics teams train on specific schemas. Layers 1–6 are in this vault. 7–9 are EgoVista-class partner work — we do not run that GPU pipeline.
| # | Layer |
|---|---|
| 1 | 2D hand pose |
| 2 | Contact timing |
| 3 | Action labels |
| 4 | Narration track |
| 5 | Face blur |
| 6 | Metadata + QA score |
| 7 | 3D hand posepartner |
| 8 | Depth + hand-object segmentationpartner |
| 9 | Camera intrinsicspartner |
Pixel files stay on your device. This vault stores the episode card. Read the footage licence.