Pricing
A fixed price for every accepted item.
We bill only work a person has accepted, at a published price, and say what the expert receives. Nothing is billed for answers that are returned or rejected.
Per accepted item
| Item | What you receive | Expert's time | Price | To the expert |
|---|---|---|---|---|
| Verified coding or agent task | Problem statement, starting code, reference solution and hidden tests, proven in our sandbox, with pass@k where you allow it | about 2 hours, senior engineer | $320 | $200 (63%) |
| Environment task | Instruction, starting data, grader checks and a reference solution, audited against reward hacks | about 2 hours, business-ops expert | $320 | $200 (63%) |
| Ideal answer | An expert-written answer to your prompt, reviewed | 30 to 60 minutes | $95 | $60 (63%) |
| Rubric grade | A 1–5 score per criterion, an overall score and a comment, reviewed | 15 to 20 minutes | $40 | $25 (63%) |
| Preference pair | The better of two answers, how strongly, and why, reviewed | 10 to 15 minutes | $30 | $18 (60%) |
Per expert hour
For work that doesn't come in items, such as designing an environment: the time measured on our server on accepted answers, capped per answer at a limit we agree with you.
| Field | Price an hour | To the expert |
|---|---|---|
| Coding and AI agents | $160 | $100 (63%) |
| Finance, legal, medicine and writing | $145 | $90 (62%) |
Included, never billed
- Reviewer time: a person checks every answer before you see it.
- The second look, when a decision on an answer is questioned.
- The experts' training and calibration time.
- Hidden known-answer tasks that measure accuracy.
- Answers that are returned for changes or rejected.
How billing works
- Accepted means a person approved it; for coding and environment tasks, the sandbox's checks passed too.
- A new project starts with a small first batch, billed once you approve it.
- A pilot's deposit is 40% of its estimate, invoiced on signing and credited against delivered work.
- No minimum. For large orders, ask us for volume pricing.
- A project keeps the price it started with: a change to this sheet applies to new projects only.
If an answer falls short
- Return any delivered answer within 14 days. A reviewer other than the one who approved it decides within two business days: it's replaced and credited on your next invoice, revised by its expert, or stands, with a written reason.
- Twice back: if a person confirms a returned answer was AI-written or copied, you're credited twice its price.
- How returns work
With every delivery
- A quality report and a datasheet, week by week.
- Exports as JSONL, preference pairs or Harbor tasks, and a read-only API.
- Where the data lives and who can see it: our security overview.
Questions about a quote: support@canonset.com; a person answers within one business day.
Ready to judge the data on your own tasks?
Start with a 50–200 task pilot, and pay for answers that pass review. A person replies within one business day.