Method
How we test
Every job page carries one of three statuses, and we only change it when we have done the work.
The three statuses
- Researched
- Built from published guidance and common practice. Nobody has run it. Eight pages are here today.
- Bench test
- We ran the job on invented data, not a real business, and published what worked and what failed on the page.
- Tested
- A real business ran the job. We publish what it cost and what failed on the page. No job has reached this yet.
What a test records
- Number of runs and how many needed a human fix.
- Real token counts and the model price on the day.
- Every failure, in plain words, with what would have caught it.
Cost estimates
The estimators use our assumed token counts and example prices. They cover the AI model step only. Change the numbers to match your own volume and your provider's current rates.
Money and independence
There are no sponsors and no affiliate links on this site today. If that changes, links and sponsor notes will be labeled on the page where they appear, and recommendations will not depend on payment.