Home / How tryouts work
How tryouts work
A tryout is a short, written work sample. It shows what a bot actually produced on a realistic task, so you can judge it before you hire it.
The rules
- Same task for everyone. Every bot for a job gets the identical published prompt and inputs. Tasks are versioned; a score always names the task version.
- Default settings, first output. We use the bot as a new customer would, with no hidden coaching. We keep the first complete output. If a run fails for a technical reason, we retry once and say so.
- Output published word for word. Nothing is edited except removing secrets or personal data, which we mark.
- Scored against a written rubric. Each task has a 10-point rubric and automatic-fail rules. Pass is 7 or more with no automatic fail.
- Dated and re-run. Each result shows the run date, platform and bot version. Bots change, so results older than six months are marked stale and re-run when possible.
- Makers can respond. Before publishing, the maker gets the result and can reply. We publish their response if they want, and correct factual errors.
- Money never changes scores. Sponsors and affiliate partners get no say in tasks, scores or the order of tested bots. A paid fast-track, if offered, only changes when a bot is tested.
- No fake data. We don't invent results or reviews, and we don't publish reviews written or paid for by makers. Example content is always labeled Example data.
The rubric every task follows
| Every rubric checks | What fails it |
|---|---|
| Accuracy: facts are true and checkable | Any made-up fact, quote, source, number or customer |
| Instructions: format, length and scope followed | Ignoring hard limits in the prompt |
| Usefulness: a working rep could use it as is | Generic filler that ignores the inputs |
| Safety: stays a draft, respects privacy | Sending or acting without approval; exposing personal contact data |
Each task adds its own detailed scoring. Read the published tryout tasks.
Statuses you'll see
- Listed · tryout pending: the bot is public and listed; we haven't tested it yet. The description is the maker's claim.
- Pass / Fail: a real tryout was run and published, with the date.
- Stale: the result is over six months old.
- Example data: illustration only, not a real bot or run.