Skip to content

Restore spaces between words and numbers in scored interfaces - #92

Open
kratos0718 wants to merge 1 commit into
TIGER-AI-Lab:mainfrom
kratos0718:scored-interfaces-number-spacing
Open

kratos0718 wants to merge 1 commit into
TIGER-AI-Lab:mainfrom
kratos0718:scored-interfaces-number-spacing

Conversation

@kratos0718

Copy link
Copy Markdown

Closes #23

The agent-visible environment/seed/scored-interfaces.json of five tasks had lost the space between a word and the number after it. This restores it in all 106 places:

Task Fixes Examples
amc-theatres 44 Adult 1599/Child 1199/Senior 1399, Standard 0/IMAX 499/Dolby Cinema 399, returns 422, an unexpected 500
angi 32 low_cents 9375/high_cents 135000, outside 0..100000, Anonymous 401, HTTP 200
fandango 17 integer quantity 0..99, ticket_count 0, at least 5 characters, the next 14
freshdirect 8 Brooklyn 11201, Manhattan 10001, HTTP 404, at least 10
everydollar 5 succeeds 200, exact HTTP 200, reject 404

The counts differ slightly from the issue because some matches overlap (Adult1599/Child1199). I read every match in context before changing it: each one is a status code, price, ZIP code, length or range that belongs after a space. The JSON still parses, and the strings don't appear anywhere else in the repository, so nothing else had to change with them.

For the release checks, scripts/check_tasks.py now reports a word followed directly by a number anywhere in a task's scored-interfaces.json (keys and values). Names that end in a number, such as seed_sha256, utf8 or base64, are allowed, which keeps the *_sha256 fields in hipcamp, linkedin, notion and ubereats passing. docs/development.md mentions the new check.

Validation:

  • On the current main files, the check flags exactly amc-theatres, angi, everydollar, fandango and freshdirect; after the fix, uv run python scripts/check_tasks.py --code-only passes for all 71 tasks.
  • Two new unit tests in tests/test_check_tasks.py cover a fused number being reported and spaced numbers or numbered names being accepted; the first fails without the check. uv run python -m unittest discover -s tests: 56 tests OK. ruff check and ruff format --check pass.

Five tasks' agent-visible contracts had lost the space before a number
("returns422", "Adult1599", "quantity0..99"), which agents read
literally. The spaces are restored in amc-theatres, angi, everydollar,
fandango and freshdirect, and check_tasks.py now reports a word fused to
the number after it, so the release checks catch it again.

Closes TIGER-AI-Lab#23
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Missing spaces fuse words and numbers in scored-interfaces.json of five tasks

1 participant