Help center
How Canonset works for experts and AI teams: pay, reviews, the ID check, the API, exports and the datasheet.
For experts
Pay, returned answers, the ID check, applying again, your quality score, and original work.
- How experts get paid
What Canonset pays for, when the money goes out, how it reaches your bank, and what happens to pay when an answer is returned, not accepted or replaced after delivery.
- When an answer is returned or not accepted
What a return means and how to fix and resend, what not accepted means for your pay and score, how to ask for a second look, the pre-check before you send, and the warning that comes before any deactivation for quality.
- The ID check
Why Canonset verifies identities before paid work, what the check collects, how consent works, where the data is kept and for how long, the video call alternative, and what happens when a check can't be settled.
- Applying again after a rejection or a waitlist
When you can apply to Canonset again after a decline or a deactivation, how the waitlist for states on hold works, what a new application involves, and what changes your chances.
- Your quality score and what moves it
How Canonset works out an expert's quality score from review scores and known-answer tasks, what returns, rejections, second looks and client returns do to it, and what it changes about the work you get.
- AI tools and original work
Canonset's rule on AI-written work, how answers are checked, what the pre-check warns you about before you send, what a confirmed breach costs, and why clients need original work.
For AI teams
The API, webhooks, the export formats, the datasheet, and returning a delivered answer.
- The Canonset API
Read-only API keys, the three endpoints (projects, export, datasheet), the rate limit, the error shape, and the audit log every export goes into.
- Webhooks and the signature
Which events Canonset's delivery webhook sends, what each carries, how to verify the Canonset-Signature header, how long you have to answer, and how failed deliveries are retried.
- Export formats, raw JSONL, preference pairs and Harbor tasks
What each line of the raw JSONL export contains for each task type, how DPO preference pairs are made and the majority rule behind them, the layout of a Harbor package and how to run it, and how to pull only new answers with since.
- The datasheet that comes with every delivery
What the Markdown datasheet says about a delivery, section by section (what it is, where it came from, how it was made and checked, the quality figures, the rights in it and its retention), and how to fetch it in the app or by API.
- Returning a delivered answer
How a client returns an approved answer within 14 days of delivery, the reasons to choose from, what a second reviewer decides and what each outcome means for the data and the invoice, and where to see each return's state.
Not answered here? Email support@canonset.com: a person answers within one business day.