[wuyuxin38324]

Following the discussion in #14, this PR corrects only confirmed errors in existing task metadata and one mathematical expression in the rubric. The original source investigation and patch were contributed by @wuyuxin38324; the maintainer follow-up narrows the changes to the minimum needed.

Task	Correction
Earth_003	Describe the packaged input as a 1-degree (2, 70, 181, 360) sample and 006.nc as a single +6 h forecast, instead of describing the files as full-resolution inputs and a 15-day sequence.
Life_002	Identify 7XG4 as the nine-chain structural hit and 6N40 as a one-chain MmpL3 structure, rather than treating 6N40 as the multi-chain target from the published example.
Physics_001	Remove the claim that the supplied arrays directly support all quantitative fits when several response lengths do not match the supplied coordinate lengths. Correct the log-axis keyword from log(T_c-T) to log(T/T_c), as implied by the existing D_s(T) = D_s(0)[1 - (T/T_c)^n] formula.

No new task files are added. Numerical arrays, PDB coordinates, task goals, and rubric weights are unchanged. The exact Life_002 reference pairing and the missing Physics_001 coordinate mappings remain unresolved; this PR does not infer replacements for them.

Validation: the modified JSON files parse, declared input paths exist, task_info.json changes are limited to five descriptions, and the checklist change is limited to one log-axis keyword. The repository smoke workflow also runs on this PR.