[Benchmark Backfill] Clarify ArcAGI1-Image naming and scope

[Luodian]

Candidate from LMM-299 is ambiguous. Confirm whether ArcAGI1-Image maps to ARC-AGI-1 rendered-image evaluation or a separate benchmark; define exact task scope for implementation.

[Luodian]

Research update for ArcAGI1-Image scope:\n\n- Official ARC-AGI-1 is distributed as grid JSON (not image benchmark): https://github.com/fchollet/ARC-AGI\n- I could not find an official/public Hugging Face dataset named ArcAGI1-Image.\n- is not publicly accessible (HF returns HTTP 401) and is not in the public dataset listing.\n- Public image-converted ARC-AGI-1 datasets do exist, but they appear ad-hoc/community repackagings with unclear benchmark standardization (e.g., , ).\n\nConclusion: ArcAGI1-Image is currently ambiguous and lacks a canonical public benchmark source.\n\nRecommendation: close this ticket for now (or re-open when we have an agreed canonical dataset + scoring protocol).

[Luodian]

Correction note (dataset IDs rendered incorrectly in my prior comment due shell quoting).

Research update for ArcAGI1-Image scope:

• Official ARC-AGI-1 is distributed as grid JSON (not image benchmark): https://github.com/fchollet/ARC-AGI
• I could not find an official/public Hugging Face dataset named ArcAGI1-Image.
• barc0/200_exclusive_arc-agi-1 is not publicly accessible (HF returns HTTP 401) and is not in the public barc0 dataset listing.
• Public image-converted ARC-AGI-1 datasets do exist, but they appear ad-hoc/community repackagings with unclear benchmark standardization (e.g., mertaylin/arc-agi-images, ayjays132/ARC_AGI_V1_ULTRA).

Conclusion: ArcAGI1-Image is currently ambiguous and lacks a canonical public benchmark source.

Recommendation: close this ticket for now (or re-open when we have an agreed canonical dataset + scoring protocol).

[Luodian]

Resolved via PR #1190 (merged). ARC-AGI-1 task added.