| --- |
| license: cc-by-nc-4.0 |
| task_categories: |
| - question-answering |
| language: |
| - en |
| tags: |
| - reasoning |
| - mathematics |
| - physics |
| - chemistry |
| - jee |
| - failure-analysis |
| - contamination-aware |
| size_categories: |
| - n<1K |
| extra_gated_prompt: "Please share your details to instantly access this sample dataset. Your request is auto-approved." |
| extra_gated_fields: |
| Name: text |
| Email: text |
| Organization: text |
| Intended use: text |
| --- |
| |
| # NULLSPACE |
|
|
| **267 JEE questions frontier models get wrong — every answer key independently verified, every row tagged for contamination.** This is a 10-row sample. Access is auto-approved — fill in your details to view the data. |
|
|
| > Full dataset (267 questions · 77 held out): [Nalandadata/NULLSPACE](https://huggingface.co/datasets/Nalandadata/NULLSPACE) — requires access request. |
|
|
| NULLSPACE is a failure-only benchmark. A question enters the set only when multiple frontier models independently get it wrong and the answer key has been re-verified. Each row carries a `do_not_score` field listing models that must be excluded when scoring — because those models' failures were used to build the set, scoring them on it measures nothing. Clean accuracy (67.3%) is higher than overall accuracy (62.2%) because contaminated rows are the ones models specifically struggled with. |
|
|
| --- |
|
|
| ## Sample coverage |
|
|
| | item_id | Subject | Answer type | Figures | |
| |---|---|---|---| |
| | STEM-0001 | Mathematics | multiple_choice | 0 | |
| | STEM-0002 | Chemistry | single_choice | 5 | |
| | STEM-0003 | Chemistry | matrix_match | 4 | |
| | STEM-0004 | Chemistry | numerical | 0 | |
| | STEM-0005 | Mathematics | multiple_choice | 0 | |
| | STEM-0006 | Physics | integer | 0 | |
| | STEM-0008 | Chemistry | multiple_choice | 5 | |
| | STEM-0010 | Physics | single_choice | 1 | |
| | STEM-0015 | Mathematics | single_choice | 0 | |
| | STEM-0020 | Chemistry | integer | 1 | |
|
|
| ## Schema (14 columns) |
|
|
| | Field | Meaning | |
| |---|---| |
| | `item_id` | Stable public identifier, `STEM-NNNN` | |
| | `exam`, `subject`, `answer_type` | Source and question format | |
| | `question`, `options`, `column_a`/`column_b` | Question content with HTML/MathML | |
| | `answer`, `answer_key` | The key — plain string and structured object | |
| | **`do_not_score`** | **Models that must be excluded when scoring this row** | |
| | `models_failed` | Which models got this question wrong | |
| | `verdict` | `CONFIRMED_MODEL_FAILURE`, `DISPUTED_BUT_KEY_BACKED`, `DISPUTED` | |
| | `thumbnail` | First figure image (`Image` type), null for text-only | |
|
|
| ## Citation |
|
|
| ```bibtex |
| @misc{nullspace2026, |
| title = {NULLSPACE: A Failure-Only JEE Benchmark with Contamination Tracking}, |
| author = {Nalandadata}, |
| year = {2026}, |
| howpublished = {\url{https://huggingface.co/datasets/Nalandadata/NULLSPACE}}, |
| note = {267 public questions; 77 held out} |
| } |
| ``` |
|
|
| --- |
|
|
| [info@nalandadata.ai](mailto:info@nalandadata.ai) · [nalandadata.ai](https://nalandadata.ai) |
|
|
