[aadi-joshi]

Summary

leaderboard --show and --download return different leaderboards on a concluded competition, and neither says which. This adds --public so the public leaderboard is reachable from --show.

Problem

--show calls GetLeaderboard, whose override_public field the CLI never set, so it returns the final leaderboard. --download calls DownloadLeaderboard, which has no such field and returns the public one. On playground-series-s6e8:

--show                          --download
0.97176  Chris Deotte           0.97207  Chris Deotte
0.97123  MILANFX                0.97154  Changye Li
0.97120  Changye Li             0.97149  MILANFX

Ranks 2 and 3 swap, and further down the two lists hold different teams. Measurements for three competitions are on #731.

Solution

--public sets override_public on the GetLeaderboard request:

kaggle competitions leaderboard playground-series-s6e8 --show --public

  teamId  teamName      submissionDate              score
--------  ------------  --------------------------  -------
16724329  Chris Deotte  2026-08-30 23:14:31.186000  0.97207
16698321  Changye Li    2026-08-30 12:49:14.556000  0.97154
16633740  MILANFX       2026-08-31 23:53:41.770000  0.97149

which is what --download gives.

The field is only set when the flag is passed, so without it the request is what it was and --show still returns the final leaderboard. I left that default alone because it is what the competition page shows by default once a competition has concluded.

--download is unchanged, and --public does not reach it.

The help text and docs/competitions.md now say which leaderboard each flag returns. Both previously described a difference in destination only, which is what makes this easy to miss.

On --download

It still cannot return the final leaderboard. That needs either override_public added to the DownloadLeaderboard request, which is a kagglesdk and server change rather than a CLI one, or serving --download by paging GetLeaderboard. The second stays in this repo but costs one request per 200 rows, and those rows carry only team_id, team_name, submission_date and score, so the CSV would lose the Rank, SubmissionCount and TeamMemberUserNames columns the server writes today. Asked on #731 instead of picking one.

Testing

• hatch -e test run pytest tests/unit (1408 passed, 3 skipped)
• hatch run lint:all
• Live against the API: --show --public matches --download on playground-series-s6e8, --show on its own is unchanged, and on the running arc-prize-2026-arc-agi-2 both forms return the same rows
• Maintainer: please run /gcbrun when convenient

Eight new tests cover the flag setting the field, the default leaving the request alone, paging still applying alongside it, the CLI forwarding the flag, and --public not altering the download call. Seven of the eight fail against the previous code. The one that does not is the regression guard on the no-flag request, which has to pass either way. One existing parser test gained an assertion that public defaults to False.

Related

Part of #731. It does not close it, since the --download half is still open.

[aadi-joshi]

This doesn't touch --download, which still can't give you the final leaderboard. DownloadLeaderboard has no override_public field, so either it gets added there (kagglesdk + server, so not something I can do here), or --download gets rebuilt on paged GetLeaderboard calls. I could do that second one, but it's a request per 200 rows, and the rows only have team_id, team_name, submission_date and score, so the CSV would lose Rank, SubmissionCount and TeamMemberUserNames.

Please let me know which you want and I'll do it in a follow-up ! Or we leave --download public only, if the help and docs here are enough

[stevemessick]

/gcbrun

[stevemessick]

/gcbrun

[stevemessick]

Thanks! We can leave --download as-is, for now.