The task form
The task form on the HAI platform has these sections, in order:
1
Welcome
A short overview of the task, with a link to this guide.
2
Task page on GitHub
The link to the task's private GitHub page. Read it there, then come back to the form.
3
Final Task Summary
Already filled in with the draft summary. Edit it so it says, in one to three sentences, what the user asks and what any correct answer must cover.
4
Final Golden Rubric
A table of criteria, already loaded with the draft criteria. You can edit, remove or add rows (1 to 25 rows).
5
Review Outcome
Choose Accepted, Revised, Needs clarification or Unusable, and write your overall edit rationale.
6
Self-Check
A short checklist to confirm before you submit.
The columns in the rubric table
- Criterion ID: keep the draft's ID (for example c1 or p2). Give new criteria IDs like new_1, new_2.
- Description: one clear statement about the answer that a person can check on its own.
- Weight: Critical, Bonus or Penalty.
- Dimension: Correctness, Visual grounding, Instruction following or Other.
- Other dimension: appears only when you choose Other. Type a short category name, for example "clarity".
- Draft image numbers (reference): read-only. Shows which images the draft used. Check them against the images on the GitHub page.
- Image numbers needed: the images a person needs to judge this criterion, or "No image needed". Every row needs one or the other.
- Individual change log: required for every row. Write a short sentence on what you changed and why, for example "Reworded so any valid fix passes, not just one approach." or "Split from c3 so each part can be checked on its own." If you did not change it, write "No change".
What the dimensions mean
- Correctness: is the answer technically right, and would it work?
- Visual grounding: does the answer read and use the image correctly?
- Instruction following: does the answer do what the user asked?
- Other: anything else useful, such as clarity. If a draft criterion uses a dimension that is not in the list, choose Other and type that name.