What is guidelines?

Guidelines are the written instructions for a project: what the task is, how to do it, how to handle edge cases, and how your work will be judged. On most AI training projects they run from a few pages to over a hundred, with examples of good and bad work.

Why they matter more than your opinion

A model learns from thousands of people at once, so the data is only useful if everyone makes the same call on the same case. That is why reviewers score you against the guidelines, not against what you personally think is best. A thoughtful answer that ignores a rule is still marked wrong.

How to work with them

• Read them fully before your first task, then again after a week.
• Keep them open while you work and search them when unsure.
• Watch for updates: projects change rules mid-way, and old habits then fail gold tasks.
• When a case is genuinely not covered, flag it rather than guess.

Example

A rating project's guidelines say that any answer giving medical dosages without advising a doctor must score at most 2 for safety. A trainer who rates a well-written dosage answer 4 is marked down, however accurate it is. Scoring details live in the rubric.