01
Register yourself
Sign up and commit to the training programs that fit your life. Your identity anchors your digital twin — a verified, living record of the data only you can produce.
The training frontier is human
Become a model human.
AI learns from data. Good AI learns from good data — curated, consented, and human. Register, feed your digital twin, and get paid for every research set you help build.
Enrolled model humans
11,482
Across 34 countries and 6 cohort families
Sensor-hours ingested
9.7M
Multimodal, clock-aligned, quality-scored
Prompted answers
42.1M
Human context attached to raw signal
Paid to participants
$4.3M
Base rates plus licensing royalties
Help train AI to treat others how you want to be treated. The models of tomorrow are being taught today — they should be taught by people who are paid, credited, and in control.
The program
Four commitments02
Feed your twin
Use the required collection devices and software to stream real, high-fidelity data into your digital twin. Every contribution is curated, consented, and traceable to you.
03
Choose how it's used
Your data can be replicated into enhanced training sets — but only under the terms you set. You decide which research programs get access, and which never do.
04
Get compensated
You are directly paid for every research set you contribute to. When your data trains a model, you share in the value it creates.
Data catalog
Real-world signal, ready for research
Each dataset is a longitudinal, consent-tracked cohort built for a specific research gap: rights-cleared first-person task demonstrations, free-living multimodal wearables with prompted context, occupational fatigue and heat-stress labels, and high-arousal specialty populations. All records are de-identified, encrypted at rest, and released only under license.
Locked
Skilled Hands
Tradespeople, technicians, makers at work
- n = 1,800
- 120 days longitudinal
- Egocentric video, IMU, audio, tool telemetry
- Task narration, step labels, error & outcome logs
Composition
- Machining, assembly, repair, culinary, lab work
- Novice to master skill tiers
- Successes and failures both recorded
Locked
Daily Signal
Free-living adults across income strata
- n = 5,000
- 180 days longitudinal
- HR, HRV, sleep, activity, phone passives
- 3–5 prompted micro-surveys per day
Composition
- Income: <35k, 35–75k, 75–150k, >150k
- Urban, suburban, rural balanced
- Age 21–65, employed & retired
Locked
Factory Floor
Manufacturing workers, three shifts
- n = 2,400
- 90 days longitudinal
- Accelerometer, HR, SpO₂, heat strain, noise
- 6 prompted Q&A blocks per shift
Composition
- Assembly, machining, logistics roles
- Fatigue & heat-stress labels, day/night shifts
- Age 22–58, 54% male / 46% female
Locked
Extreme Cohorts
Climbers, parachutists, high-arousal sport
- n = 1,750
- 120 days longitudinal
- HR, HRV, EDA, IMU, GPS, barometer
- Pre/post event stress + reflection logs
Composition
- Indoor to alpine, student to instructor
- Graded exposure & performance tiers
- Jump/ascent count 10–4,000+
Custom research
Commission your own study
Define your population, sample size, sensor package, and study duration. We recruit the cohort, manage consent and devices, collect the signal, and deliver a clean, documented dataset ready for model training.
- N from 100 to 10,000+
- 30 to 365-day protocols
- Wearables, phones, ambient sensors
- Custom Q&A schedules
Data integrity
Collected with consent. Verified by hand. Stored under lock.
01
Collect
Consented participants stream wearable biometrics, egocentric task video, and phone passives through certified devices. Prompted Q&A captures context the sensors can't see.
- Certified device kits
- Prompted Q&A schedules
- Continuous consent ledger
02
Verify
Every record is traceable to a verified human. Automated signal-quality checks flag dropouts and anomalies; human reviewers audit samples and labels before release.
- Signal-quality scoring
- Identity-anchored records
- Human label audits
03
Store
Data is de-identified, encrypted at rest and in transit, and released only under license. Participants can revoke future use — revocation propagates to every downstream set.
- Encrypted at rest & in transit
- De-identified before licensing
- Revocable by the participant
Fig. 02 — Data lifecycle
Eight gated stages, one traceable record
Every sample carries its provenance from enrollment to revocation. A record cannot advance a stage until the prior gate passes.
01
Enroll
Consent ledger
02
Provision
Calibrated kit
03
Stream
Edge buffer
04
Sync
Clock align
05
Score
Signal QC
06
Label
Human audit
07
License
Scoped release
08
Revoke
Propagated
Revocation loops back to stage 01 Hash-chained audit trail Median latency 1.4 s edge → store
Instrumentation
Off-the-shelf hardware, research-grade correlation
We build on devices participants already trust, then cross-validate every channel against reference instruments so consumer signal can be used with scientific confidence.
Wrist optical + ECG
PPG 25 Hz · single-lead ECG 128 Hz
Consumer-grade wrist wearables (Garmin, Polar, Withings) validated against research reference devices before enrollment.
Chest strap ECG
R-R interval ±1 ms
Polar H10-class straps supply gold-standard HRV during exertion blocks and calibrate wrist optical drift.
Egocentric capture
1080p30 · 6-DoF IMU · stereo audio
Camera eyewear and chest mounts record first-person task demonstrations with synchronized motion and sound.
Clip-on IMU
Accel + gyro 100 Hz
Limb and tool-mounted inertial pucks resolve fine manipulation that wrist sensors alone cannot separate.
Phone passives + Q&A
GPS, screen, ambient audio level
Our app schedules prompted micro-surveys and carries context labels the sensors can never infer.
Next-gen patch (R&D)
EDA · skin temp · SpO₂ · 14-day wear
An in-development adhesive module pushing continuous autonomic signal at research fidelity in a consumer form factor.
Compensation
Paid for the signal, again for the reuse
A base rate covers your enrollment and daily contribution. A royalty pays out every time a research set containing your data is licensed again. Both are itemised in your ledger.
Monthly
Base contribution
$40 – $180 / month
Set by protocol burden: wear time, prompted answer count, and how demanding the capture package is.
Per license
Licensing royalty
12% of set revenue
Split across every participant whose records appear in a licensed research set, weighted by contribution volume and quality score.
Per protocol
Study bonuses
$50 – $900 / study
Commissioned studies, rare cohorts, and demanding capture windows carry their own negotiated rate on top of base.
Median payout latency
4 days
Quality score floor for pay
0.82
Royalty statements per year
12
Participant revocation fee
$0
Governance
Built for the compliance review, not around it
Human-subject data carries obligations. We treat consent, review, and revocation as engineering requirements so a dataset survives legal diligence as easily as a benchmark.
Consent
Granular, revocable consent
Consent is scoped per use case — model training, academic research, commercial licensing — and can be withdrawn per scope without leaving the program.
Review
Independent ethics review
Protocols involving vulnerable populations, workplace monitoring, or health inference go through external review before a single record is collected.
Privacy
Health-data grade handling
Biometric and health-adjacent signal is handled to consumer-health-data standards, with regional controls on collection, sharing, and sale.
Licensing
Auditable chain of rights
Every licensed set ships with provenance, consent scope, and a revocation feed, so buyers can prove where their training data came from.
Questions
What model humans ask first
How much time does this actually take?
Wearing the device is passive. Prompted questions run 15–90 seconds each, three to six times a day depending on the protocol you join.
Who can see my raw data?
Only the internal verification team, under audit. Licensed sets are de-identified, and identifying media like faces or audio is either excluded or redacted before release.
Can I leave?
Any time, at no cost. You can stop collecting, revoke future use of past records, and keep the payments you have already earned.
Do I need to buy hardware?
No. Certified kits are shipped to you, calibrated, and returned at the end of a protocol. If you already own a supported device we validate it instead.
How do you keep the data honest?
Records are identity-anchored, signal quality is scored automatically, and human reviewers audit samples. Synthetic or duplicated contributions are rejected and unpaid.
Who buys these datasets?
Frontier model labs, robotics and physical-AI teams, occupational health groups, and academic researchers commissioning purpose-built cohorts.
Enrollment open
Your data. Your terms. Your pay.
Join the first cohort of Model Humans. Spots in each training program are limited so every contribution stays valuable.