The homepage offers an "OTS catalog", but the only way to see it is to leave a name and email and wait for "our researcher dashboard and samples". Human Archive's strongest public proof (500 hours from 4,651 people, 259 task categories, 45 environment types) appears only on third-party dataset listings, never on its own site. Search engines can't find anything either: there is no sitemap, no robots.txt and no meta description, and a search for site:humanarchive.ai returns only unrelated "human archive" projects. Meanwhile, Digital Divide Data publishes its frame rate, pose data, hand skeleton and LeRobot v2 format openly. A researcher comparing vendors can shortlist DDD without a call, but not Human Archive.
First action
Write the HA-Ego spec sheet from the numbers already on the public sample listing and publish it at /catalog/ha-ego.
How to measure it
Catalogue sample requests from labs
- Now
- –
- Target
- 15 a month
- By
- 60 days
Steps
Draft spec sheets
One per collection: hours, contributors, tasks, environments, countries, sensors, sample rates, formats, licence.
CEO 3 days
Add the consent summary
How contributors consent, how faces are blurred and what buyers receive.
Co-founder, Operations 2 days
Build the pages
A /catalog index plus one page per collection, linked from the homepage.
Member of Technical Staff 3 days
Make it indexable
sitemap.xml, robots.txt, meta descriptions and schema.org Dataset markup; submit to Search Console.
Member of Technical Staff 1 day
Link the samples
Each page links to its gated sample and explains how access requests work.
CTO 1 day
Sharpen the form
Ask for modality, hours, task types and deadline, and route answers to the CEO.
CEO 1 day
Measure
Search impressions, catalogue views and qualified requests each week.
CEO weekly
Why it works
Build /catalog with one page per collection: egocentric RGB video, the multi-sensor rig (depth, tactile gloves, motion capture, wrist cameras) and Voice Exchange conversations. Each page lists hours, unique contributors, task categories, environment types, countries, sensors and sample rates, delivery formats, licence terms, a short consent and anonymization summary, and a link to the matching sample. Keep pricing and the researcher dashboard behind the form, but make the form ask useful questions (modality, hours, task types, deadline). Add a sitemap, robots.txt, meta descriptions and schema.org Dataset markup so search engines and dataset search can index each page. Link the catalogue from the homepage and from every launch post.
Lab buyers shortlist on specs before they talk to a vendor. Public spec pages turn anonymous research visits into qualified requests, give the CEO one link to send instead of a deck, and make the most-requested company from YC W26 findable by researchers searching for egocentric data.
Ready to use
ChecklistCatalogue page launch list
[ ] /catalog index linked from the homepage [ ] One page per collection: hours, unique contributors, task categories, environments, countries [ ] Sensors, sample rates and delivery formats [ ] Licence terms and consent/anonymization summary [ ] Link to the gated sample and how access works [ ] sitemap.xml and robots.txt [ ] Meta description and share image on every page [ ] schema.org Dataset markup [ ] Request form: modality, hours, tasks, deadline
Risks and how to handle them
Rivals copy the spec format
Specs are table stakes; the hours and sensor mix are the moat.
Publishing too much detail
Publish counts and formats; keep pricing and customer-specific collections private.
Evidence
- Your websiteHomepage (checked November 9, 2026), seven sentences in total: "Human Archive is a data and deployments infrastructure company for Physical AI." "Human Archive is currently working alongside leading AI labs to scale human video/sensor data and deploy robotics." "If you're interested in our OTS catalog or a robot deployment, click here or request time here." The links go to a contact form, a 30-minute booking page, an X post and the legal pages. There is no catalogue page, no link to the public sample dataset, no link to the iPhone app and no careers link.Source
- Your websiteContact page: "Get in touch" and "Drop your details and we'll send over our researcher dashboard and samples." The only fields are name, email and notes.Source
- Dataset listingHA-Ego-Samples, Human Archive's public sample: "500 hours of footage from 4,651 unique individuals", 9,343 clips across 259 task categories and 45 environment types; RGB video only (MP4); Apache-2.0 licence; access by request on Hugging Face; 163 downloads shown on the listing.Source
- Your websiteCrawl check (November 9, 2026): /sitemap.xml and /robots.txt both return 404, and a search for site:humanarchive.ai returned no pages from the domain, only unrelated projects with "human archive" in their names. The domain was registered in October 2025.Source
- Competitor researchDigital Divide Data's egocentric dataset page publishes its specs: rectified stereo video above 1080p at 60 fps, per-frame 6DoF head pose, a 21 to 26-joint hand skeleton per hand, IMU, delivery in HDF5 and LeRobot v2, and the capture headsets used, with "Access Dataset" buttons.Source