Uncensored AI for citizen science projects
Project designs and data notes good enough to be used
Citizen science succeeds or fails at the protocol: if volunteers cannot follow it the same way twice, the dataset is noise with a website. Writing a protocol that survives contact with the public is a real skill.
Why the mainstream APIs fail here
Protocols written by experts assume expert noticing. Data cleaning swallows the project afterwards because observations arrived in nine formats with unknown in the species column, and the model that could help hedges on the invasive-species and disease-watch projects, which are half the useful ones.
How Wild West API is used for this
Draft the protocol with the model playing a confused volunteer: anything it misreads, real volunteers will misread worse. Turn observations into clean records with the same vocabulary rules throughout. Ask what bias the design has; reviewers always ask.
A working loop
- Play the protocol against a confused reader before launch.
- One vocabulary for species and habitats, everywhere, from day one.
- State the design bias in the methods; it will be found anyway.
Details are on the privacy page. Per-model prices are on /models and the pricing model is on /pricing.