Marketplace
Data you can license, and data we can collect for you.
Each entry describes a kind of data we record. Every dataset ships with a complete record of where it came from.
Domain
All 21 kinds of data
- Behavior
Spreadsheet and finance workflows
Month-end close, reconciliation and audit preparation in Excel and ERP systems, with the corrections and the final result.
screen video, input events, outcomes
- Behavior
Documents and slides
Authoring sessions in Word and PowerPoint, from first draft through revisions to the finished file.
screen video, revisions, final files
- Behavior
Creative tool workflows
Design work in Figma and Illustrator, including the review cycles between versions.
click streams, tool use, review cycles
- Behavior
Software engineering sessions
IDE, terminal and browser activity during real feature work, linked to the commits it produced.
editor events, terminal, commits
- Behavior
Legal and administrative review
Contract markup, clause comparison and approval chains, recorded as the work happens.
input events, markup, decisions
- Behavior
Custom software cohorts
A consenting team recorded on the exact software your agent needs to use.
your stack, your tasks
- Skill
Textile and garment production
First-person video of sewing, finishing and quality checks on working lines, with hand pose and process steps.
egocentric video, hand pose, process steps
- Skill
Loom setup and fault recovery
Overhead and wrist cameras on loom threading, setup and recovery from faults.
multi-camera video, fault recovery, tool events
- Skill
Metal fabrication
Cutting, welding and fit-up from the worker’s point of view, with protective equipment regions anonymized.
egocentric video, tool events, safety events
- Skill
Field trades
Plumbing and electrical work on site, from diagnosis to repair and verification, recorded against your task list.
egocentric video, diagnosis steps, instrument readings
- Judgment
Science preference pairs
Graduate-level problems with ranked model responses, written rationales and step-by-step corrections.
preference pairs, rationales, error notes
- Judgment
Legal reasoning evaluations
Practicing attorneys scoring contract analysis and case reasoning against a rubric, with written justifications.
rubric scores, justifications
- Judgment
Code review preferences
Senior engineers ranking model-written patches for correctness, safety and style, with inline comments.
rankings, inline comments
- Judgment
Accounting and audit evaluations
Accountants scoring model answers on reconciliation, disclosure and audit reasoning.
rubric scores, gold tasks
- Judgment
Clinical decision panels
Physicians producing preference data on diagnoses and treatment plans, built to your evaluation framework.
physician panels, preference pairs
- Judgment
Red-team panels
Domain experts probing for confident errors and unsafe advice in high-stakes fields.
adversarial prompts, failure reports
- Perception
CT reads with reasoning
De-identified studies paired with the radiologist’s impression, differentials and follow-up decision at the time of the read.
DICOM, reads, differentials
- Perception
Radiographs with reports
X-rays across orthopedics, chest and emergency care, with structured reports and demographic context.
X-ray, structured reports
- Perception
Ultrasound with sonographer notes
Full sweep recordings with probe settings, measurements and annotated findings.
ultrasound video, measurements, notes
- Perception
Hospital operations events
De-identified workflow events from working hospitals: orders, handoffs and escalations.
event streams
- Perception
Prospective clinical cohorts
Collected against your protocol, from modality and population to annotation schema, under ethics review.
protocol-defined
Start with a conversation.
A short scoping call is enough to see what your model is missing and how we can help.
Every delivery includes a signed record of where the data came from.
How provenance works