gpqa-d: official 198 questions; mmlu-pro-subset: pinned official source

Previous RLLab/eval-set revision: acbeda874c86be821a4927e4c1f824f63980630a
aime24: 30 -> 30 questions; 0 prompts changed; all gold strings and row order retained.
aime25: 30 -> 30 questions; 0 prompts changed; all gold strings and row order retained.
gpqa-d: 197 -> 198 questions; 1 added; of the 197 retained questions, 8 are byte-identical, 8 identical up to whitespace, and 49 keep their gold letter.
mmlu-pro-subset: 10681 -> 10681 questions; 0 added; of the 10681 retained questions, 10344 are byte-identical, 10681 identical up to whitespace, and 10681 keep their gold letter.

Sources:
- [MathArena/aime_2024_I](https://huggingface.co/datasets/MathArena/aime_2024_I/tree/ea5b061c3e8039dc9858defaafc407d04b995e9f)
- [MathArena/aime_2024_II](https://huggingface.co/datasets/MathArena/aime_2024_II/tree/29d5d31e9b46e215fc24d9b2a3047506823dd101)
- [MathArena/aime_2025](https://huggingface.co/datasets/MathArena/aime_2025/tree/c94da77eb22bbd6439e62a323bec18493a421302)

Preprocessing: remove Asymptote blocks (`[asy]`, `<asy>` and the LaTeX `asy` environment) and TikZ `tikzpicture` environments. Preserve all prose, mathematical formulas and tables outside those blocks. Collapse excess blank lines, trim outer whitespace, and append the step-by-step / boxed-answer instruction. Question order and gold formatting are unchanged.

Previous AIME mirrors used by AetherEval:

- `aime24`: [HuggingFaceH4/aime_2024](https://huggingface.co/datasets/HuggingFaceH4/aime_2024).
- `aime25`: [yentinglin/aime_2025](https://huggingface.co/datasets/yentinglin/aime_2025).

The pre-migration `eval-set` configs matched those AetherEval copies in all 30 questions per year, row order and answers, ignoring whitespace and the appended math instruction. The original `eval-set` import provenance was not recorded, so these are identified as matching previous mirrors, not a verified import history.

## Previous data repairs

The original commits contain the complete row-level changes:

- **OlympiadBench (675 → 674 questions):** previous content matched [knoveleng/OlympiadBench](https://huggingface.co/datasets/knoveleng/OlympiadBench), an older mirror; the original import provenance was not recorded. Rebuilt from `Hothan/OlympiadBench`, config `OE_TO_maths_en_COMP`. Updated 15 relay question transcriptions from the official release, removed one question removed upstream, and followed upstream's degree-unit formatting for one gold answer. [Full commit](https://huggingface.co/datasets/RLLab/eval-set/commit/93bfada9b7e597586ab6cdf68d98ff88ae4a3f3a).
- **MinervaMath (272 questions, unchanged):** source [knoveleng/Minerva-Math](https://huggingface.co/datasets/knoveleng/Minerva-Math/tree/93d86e1f779d84c7393587c2cdbf6c7511591f95). The pre-repair `RLLab/eval-set` / `minervamath` rows match this source in all 272 questions, solution texts and row order, apart from the appended math instruction. Corrected the notation of 62 gold answers (58 scientific-notation and 4 symbolic rewrites) so `math-verify` preserves their intended values. Question text, order and mathematical answers are unchanged. [Full commit](https://huggingface.co/datasets/RLLab/eval-set/commit/c774e009046b9173716066989a392d0851ea9e4a).

## Multiple-choice sources

- **`gpqa-d` (198 questions):** official GPQA Diamond, from the [simple-evals CSV](https://openaipublic.blob.core.windows.net/simple-evals/gpqa_diamond.csv) (sha256 `41d1213cd7a4998605a26c2798500652572007161b3a92817ba46b35befcd305`). The previous config held 197 of these questions; its import provenance and option shuffle were not recorded. Options now follow the permutation of openai/simple-evals `gpqa_eval.py` (`random.Random(0)`, `rng.sample(range(4), 4)` per question in CSV order), so gold letters differ from the previous revision. Question and option text is stripped of outer whitespace.
- **`mmlu-pro-subset` (10,681 questions):** [TIGER-Lab/MMLU-Pro](https://huggingface.co/datasets/TIGER-Lab/MMLU-Pro/tree/b189ec765aa7ed75c8acfea42df31fdae71f97be) test split without the `math` category (1,351 questions), in official order with official text. Gold letters are unchanged.

Both use the prompt `The following are multiple choice questions (with answers)[ about <category>]. Think step by step ... \boxed{}`, then `Question:`, a blank line, and `Options:` lettered from A.

README also documents the previous OlympiadBench/MinervaMath repairs; their data is not rewritten.

Prompt changes require new generations; rescoring old answers is not equivalent.

README.md

@@ -44,10 +44,10 @@ dataset_info:
     dtype: string
   splits:
   - name: train
-    num_bytes: 163279
-    num_examples: 197
-  download_size: 74943
-  dataset_size: 163279
+    num_bytes: 169102
+    num_examples: 198
+  download_size: 77826
+  dataset_size: 169102
 - config_name: math500
   features:
   - name: problem
@@ -80,10 +80,10 @@ dataset_info:
     dtype: string
   splits:
   - name: train
-    num_bytes: 10035691
+    num_bytes: 10035337
     num_examples: 10681
-  download_size: 4011987
-  dataset_size: 10035691
+  download_size: 3958677
+  dataset_size: 10035337
 - config_name: olympiadbench
   features:
   - name: problem
@@ -163,3 +163,12 @@ The original commits contain the complete row-level changes:
 - **OlympiadBench (675 → 674 questions):** previous content matched [knoveleng/OlympiadBench](https://huggingface.co/datasets/knoveleng/OlympiadBench), an older mirror; the original import provenance was not recorded. Rebuilt from `Hothan/OlympiadBench`, config `OE_TO_maths_en_COMP`. Updated 15 relay question transcriptions from the official release, removed one question removed upstream, and followed upstream's degree-unit formatting for one gold answer. [Full commit](https://huggingface.co/datasets/RLLab/eval-set/commit/93bfada9b7e597586ab6cdf68d98ff88ae4a3f3a).
 - **MinervaMath (272 questions, unchanged):** source [knoveleng/Minerva-Math](https://huggingface.co/datasets/knoveleng/Minerva-Math/tree/93d86e1f779d84c7393587c2cdbf6c7511591f95). The pre-repair `RLLab/eval-set` / `minervamath` rows match this source in all 272 questions, solution texts and row order, apart from the appended math instruction. Corrected the notation of 62 gold answers (58 scientific-notation and 4 symbolic rewrites) so `math-verify` preserves their intended values. Question text, order and mathematical answers are unchanged. [Full commit](https://huggingface.co/datasets/RLLab/eval-set/commit/c774e009046b9173716066989a392d0851ea9e4a).
 <!-- /eval-set-repairs -->
+
+<!-- mcq-sources -->
+## Multiple-choice sources
+
+- **`gpqa-d` (198 questions):** official GPQA Diamond, from the [simple-evals CSV](https://openaipublic.blob.core.windows.net/simple-evals/gpqa_diamond.csv) (sha256 `41d1213cd7a4998605a26c2798500652572007161b3a92817ba46b35befcd305`). The previous config held 197 of these questions; its import provenance and option shuffle were not recorded. Options now follow the permutation of openai/simple-evals `gpqa_eval.py` (`random.Random(0)`, `rng.sample(range(4), 4)` per question in CSV order), so gold letters differ from the previous revision. Question and option text is stripped of outer whitespace.
+- **`mmlu-pro-subset` (10,681 questions):** [TIGER-Lab/MMLU-Pro](https://huggingface.co/datasets/TIGER-Lab/MMLU-Pro/tree/b189ec765aa7ed75c8acfea42df31fdae71f97be) test split without the `math` category (1,351 questions), in official order with official text. Gold letters are unchanged.
+
+Both use the prompt `The following are multiple choice questions (with answers)[ about <category>]. Think step by step ... \boxed{}`, then `Question:`, a blank line, and `Options:` lettered from A.
+<!-- /mcq-sources -->

gpqa-d/train-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8864b89cf301287220afd31b164834fae41c1ac37e93a8886bd00e43020da5be
-size 74943
+oid sha256:fcb823d0972ffdcf6afa13dfe885fc925116f34aa68889c6f02e96e016dbb82e
+size 77826

mmlu-pro-subset/train-00000-of-00001.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9076ba56f85ea0f94847848d14eb72eb064636a62deeb4aed4be71a925460531
-size 4011987
+oid sha256:4606a58e54cb072547ec1b8ab841a06ab6ec43f4c3cfa2fcabf52ce028904be7
+size 3958677