DataWork Watch dataset - documentation
The dataset is a live, sourced tracker of paid AI data-collection
programs. Machine formats: JSON (ETag +
Last-Modified, CC-BY-4.0) and CSV.
Change feed: Atom. Per-work-type
feeds: /datawork/type/<wt>/feed.xml (+.json,
JSON Feed) where a type has ≥3 programs. Developer API:
/api. Weekly summary:
report.
Quick answers
What formats is the DataWork Watch dataset available in?
JSON at https://bikoosh.com/datawork/data.json (ETag and Last-Modified supported) and CSV at https://bikoosh.com/datawork/data.csv, both CC-BY-4.0. Changes are published as an Atom feed at https://bikoosh.com/datawork/changes.xml, and each work type with three or more programs has its own Atom and JSON Feed.
What does the application_status field mean?
One of open, waitlist, closed or unknown. It is set only by an explicit phrase match on the program's own page or by a manual, sourced review. It is never inferred.
What does the legit_status field mean?
In the public dataset it is one of two values: verified, or unreviewed_public for everything else. A verified record cites at least two independent sources on its page. Verdicts are manual and never change automatically.
What is the difference between pay_claimed and pay_reported?
pay_claimed is the pay the operator states on its own pages; pay_reported is pay that workers have independently reported. Each carries its own source, or is null when no sourced figure exists.
How often is the DataWork Watch dataset updated?
Each program's own page is checked at most once a day, with one fetch per domain per day, the honest BikooshBot user-agent, and robots.txt respected. Ambiguous changes go to human review; a status never changes from ambiguity.
How do I cite DataWork Watch?
Cite it as: DataWork Watch (https://bikoosh.com/datawork), retrieved on the date you accessed it, live dataset of paid AI data-collection programs. Link the hub or a program page. The dataset is CC-BY-4.0, so attribution is the only requirement.
Field dictionary
| field | meaning |
|---|---|
| slug | stable identifier; program page at /datawork/<slug> |
| application_status | open · waitlist · closed · unknown - set only by an explicit phrase match on the program's own page or a manual sourced review; never inferred |
| legit_status | verified · unreviewed_public - the public dataset carries these two only. A verdict of verified cites at least two independent sources on the page; everything else is published as unreviewed_public, and the screening notes behind that stay internal. |
| work_types | annotation, rating, expert, coding, voice, video_recording, egocentric, home_scan, transcription, other |
| countries_mode | global · list (see countries) · unknown |
| pay_claimed / pay_reported | operator-claimed vs worker-reported pay; each carries its own source or is null |
| verified_at | last time our checker re-verified the program's own page |
| change_log | full dated history of every field change with a note |
Update cadence
Each program's own page is checked at most once a day (politeness cap: one fetch per domain per day, honest BikooshBot user-agent, robots.txt respected). Ambiguous changes go to human review - statuses never change from ambiguity.
Use this data in Google Sheets
=IMPORTDATA("https://bikoosh.com/datawork/data.csv?utm_source=sheets")
Live, auto-refreshing; share your sheet freely (CC-BY-4.0).
How to cite
"DataWork Watch (bikoosh.com/datawork), retrieved {{date}} - live dataset of paid AI data-collection programs." Link either the hub or a program page. The dataset is CC-BY-4.0; attribution is the only ask. Commercial license without attribution (+10x API limits): /data-license.