Fix egal_wie_pun answer key: Bruyne, not Braeune (verified against the real circulating joke; the documented punchline names the footballer directly, the tan-word is the phonetic gloss, not the written answer)

Fix middle_letters item: REH -> ZEH (answer REH was leaking into the input word list; ZEH also completes the Arm/Zeh/Ohr body-part pattern)

Fix distractor in schwarz-weiss riddle: Schachbrett -> Panda (Schachbrett was an equally valid answer under the stated criteria)

round-2 language+correctness review (1861 items)

Fluency pass: 299 adjudicated grammar/agreement fixes (incl. systematic template bugs: power_strip dative, word_half plural agreement, weekday_offset/transitive phrasing). README untouched.

Fix three_tier_liar: answer option no longer mixes an always-liar with the truth-teller (logical contradiction). 1 item corrected; generator constraint fixed at source.

hard 119->109

Fix hard subset: generated hard = robust set (missed by high+xhigh+low), 119->109

Fix prose item count 925 -> 1,861

Update dataset_info + stats for 1861 merge

Merge grips-extended: 925 -> 1861 generated reference items

Update dataset_info (column order + sizes)

Reorder columns (question, cot, answer, explanation, ..., id) + sort rows by id

Reorder schema table to match column order

Update dataset_info (column order + sizes)

Reorder columns (question, cot, answer, explanation, ..., id) + sort rows by id

Document cot column in schema table

Add cot feature to dataset_info

Upload dataset

Upload dataset

Upload dataset

card: remove orphaned funding-logo files

card: corrected hard-subset scores + grips-eval link

card: link to eval code

card: add LLM4KMU funding-agency logos

card: add LLM4KMU funding-agency logos

card: held-out leaderboard; public set as reference

Upload README.md with huggingface_hub

Upload dataset

Upload README.md with huggingface_hub

Upload dataset

Upload README.md with huggingface_hub

Upload dataset

Upload README.md with huggingface_hub

Upload dataset

initial commit