| --- |
| license: cc-by-nc-4.0 |
| language: |
| - en |
| pretty_name: KEEP-Bench |
| size_categories: |
| - 1K<n<10K |
| task_categories: |
| - summarization |
| - question-answering |
| tags: |
| - context-compression |
| - faithfulness |
| - retrieval-augmented-generation |
| - benchmark |
| - evaluation |
| configs: |
| - config_name: parallel |
| data_files: |
| - split: test |
| path: data/parallel/test.jsonl |
| - config_name: conditional |
| data_files: |
| - split: test |
| path: data/conditional/test.jsonl |
| - config_name: imbalanced_science |
| data_files: |
| - split: test |
| path: data/imbalanced_science/test.jsonl |
| - config_name: imbalanced_policy |
| data_files: |
| - split: test |
| path: data/imbalanced_policy/test.jsonl |
| extra_gated_prompt: >- |
| KEEP-Bench is released for non-commercial research on context compression. |
| The imbalanced setting contains excerpts from publicly accessible web pages |
| whose copyright remains with the original authors. By requesting access you |
| agree to use the dataset only for non-commercial research, not to |
| redistribute the web excerpts separately, and not to use the dataset as |
| training data for models evaluated on KEEP-Bench. |
| extra_gated_fields: |
| I agree to use this dataset for non-commercial research only: checkbox |
| I will not redistribute the web excerpts separately from this dataset: checkbox |
| --- |
| |
| # KEEP-Bench |
|
|
| **KEEP-Bench** (**K**ey **E**videnc**E** **P**reservation) is a diagnostic benchmark for evaluating whether context compressors preserve the evidence needed to answer a question. Instead of inferring compression quality from downstream task accuracy, KEEP-Bench annotates each case with the evidence propositions that a faithful compression must retain, and checks each proposition directly in the compressed context. |
|
|
| The benchmark contains **4,102 cases** in three settings, each targeting an evidence structure that compressors tend to break: |
|
|
| - **Parallel evidence**: the question contains underspecified mentions that admit two valid interpretations. Each interpretation leads through its own reasoning chain to a different answer, and both chains must be preserved. |
| - **Conditional evidence**: one hop of the reasoning chain holds only under an explicit qualifier, such as a period, edition, survey or version. The qualifier must be preserved together with the claim it scopes. |
| - **Imbalanced evidence**: an open-ended question has three or four valid answer perspectives. Dominant perspectives are supported by several paragraphs, while each minority perspective is supported by a single sentence. All perspectives must be preserved. |
|
|
| [Paper](https://arxiv.org/abs/2610.26827) · [Code](https://github.com/PlusLabNLP/KEEP-Bench) · [Website](https://pluslabnlp.github.io/KEEP-Bench/) |
|
|
| ## Dataset structure |
|
|
| | Setting | Config | Cases | Subsets | Avg. paragraphs | Avg. tokens | |
| |---|---|---:|---|---:|---:| |
| | Parallel | `parallel` | 900 | 400 two-hop, 300 three-hop, 200 four-hop | 22.8 | 2,799 | |
| | Conditional | `conditional` | 1,364 | 604 two-hop, 493 three-hop, 267 four-hop | 21.0 | 2,544 | |
| | Imbalanced | `imbalanced_science`, `imbalanced_policy` | 1,838 | Science: 1,010 (856 with four perspectives, 154 with three); policy: 828 (all with four perspectives) | 9.8 | 3,577 | |
|
|
| Average context length is measured with the GPT-5 tokenizer. |
|
|
| Each config has a single `test` split. KEEP-Bench is an evaluation benchmark; there is no training split. |
|
|
| ### Loading |
|
|
| Access is gated, so log in first (`hf auth login`) and accept the terms on this page. |
|
|
| ```python |
| from datasets import load_dataset |
| |
| parallel = load_dataset("uclanlp/KEEP-Bench", "parallel", split="test") |
| conditional = load_dataset("uclanlp/KEEP-Bench", "conditional", split="test") |
| science = load_dataset("uclanlp/KEEP-Bench", "imbalanced_science", split="test") |
| policy = load_dataset("uclanlp/KEEP-Bench", "imbalanced_policy", split="test") |
| ``` |
|
|
| The input to a compressor is `question` and `context`. `context` is the exact string used in our experiments; `paragraphs` holds the same content as structured records. |
|
|
| ### Fields: `parallel` |
|
|
| | Field | Type | Description | |
| |---|---|---| |
| | `id` | string | Case id, e.g. `parallel_3hop1_0001`. | |
| | `hop_type` | string | MuSiQue reasoning type: `2hop`, `3hop1`, `3hop2`, `4hop1`, `4hop2`, `4hop3`. | |
| | `hops` | int | Number of hops per chain (2, 3 or 4). | |
| | `question` | string | Question containing the underspecified mention. | |
| | `answer_1`, `answer_2` | string | Gold answer under each interpretation. | |
| | `logical_chain_1`, `logical_chain_2` | list[string] | Evidence propositions of each reasoning chain, one per hop. | |
| | `flattened_labels` | list[string] | `yes`/`no` per proposition of `logical_chain_1 + logical_chain_2`. `yes` marks the propositions that distinguish the two referents; they define Para-Rec. | |
| | `ambiguous_entities` | list[string] | Underspecified mentions in the question. | |
| | `unambiguous_entities` | list[list[string]] | For each ambiguous mention, its two explicit referents. Passed to the judge as disambiguation constraints. | |
| | `paragraphs` | list[struct] | `idx` (int), `title` (string), `paragraph_text` (string), `is_supporting` (string: `chain_1`, `chain_2` or `distractor`). | |
| | `context` | string | Paragraphs rendered as `Document [i](Title: ...)text`, one per line. | |
|
|
| ### Fields: `conditional` |
|
|
| | Field | Type | Description | |
| |---|---|---| |
| | `id` | string | Case id, e.g. `conditional_2hop_0001`. | |
| | `hop_type`, `hops` | string, int | As in `parallel`. | |
| | `question_group` | string | Cases sharing the same question, usually paired with different qualifiers (536 groups have two cases). | |
| | `question` | string | Question. | |
| | `answer` | string | Gold answer, stated together with its qualifier. | |
| | `key` | string | The qualifier, e.g. `In the 1990s`. | |
| | `logical_chain` | list[string] | Evidence propositions, one per hop. Exactly one proposition contains `key`; it is the qualified proposition. | |
| | `disambiguation_proposition` | string | States that the qualified relation differs across qualifiers. It explains the scope of the qualifier to the judge and is **not** an evaluated proposition. | |
| | `paragraphs` | list[struct] | `idx`, `title`, `paragraph_text`, `is_supporting` (string: `supporting`, `disambiguating` or `distractor`). The `disambiguating` paragraph shows that the relation is not valid without the qualifier. | |
| | `context` | string | Rendered as in `parallel`. | |
|
|
| ### Fields: `imbalanced_science` and `imbalanced_policy` |
|
|
| | Field | Type | Description | |
| |---|---|---| |
| | `id` | string | Case id, e.g. `science_1major_3minor_0001` (1 dominant, 3 minority perspectives). | |
| | `domain` | string | Scientific field (science) or policy scenario (policy). | |
| | `num_dominant`, `num_minority` | int | Number of dominant and minority perspectives. | |
| | `question_group` | string | Cases sharing the same question with different dominant/minority configurations. | |
| | `question` | string | Open-ended question. | |
| | `perspectives` | list[string] | Answer perspectives, i.e. the evidence propositions. Policy perspectives are prefixed with their jurisdiction. | |
| | `perspective_roles` | list[string] | `dominant` or `minority` for each perspective, aligned with `perspectives`. | |
| | `paragraphs` | list[struct] | `idx` (int), `text` (string), `label` (string). | |
| | `context` | string | Paragraphs rendered as `Document [i]text`, one per line. | |
|
|
| Paragraph `label` values refer to positions in `perspectives`: |
|
|
| - `major_<i>_supporting`: a chunk supporting dominant perspective `i`. |
| - `minor_<i>_supporting`: the chunk hosting the single sentence that supports minority perspective `i`. The host is a retrieved chunk into which that sentence was inserted, so the rest of the chunk may support another perspective. |
| - `unsupporting`: a chunk that supports none of the perspectives (123 chunks in 95 science cases). |
|
|
| ## Evaluation |
|
|
| The evaluation protocol, metrics (All-Rec, Para-Rec and Cond-Rec) and judge prompts are described in the paper and released in the code repository. |
|
|
| Email addresses and phone numbers in the web excerpts have been replaced with `[EMAIL]` and `[PHONE]`; these strings are not part of any evidence proposition. |
|
|
| ## License and terms of use |
|
|
| KEEP-Bench is released under [CC BY-NC 4.0](https://creativecommons.org/licenses/by-nc/4.0/). |
|
|
| - The parallel and conditional settings are derived from [MuSiQue](https://github.com/StonyBrookNLP/musique) (CC BY 4.0). |
| - Questions, evidence propositions, annotations and rewritten passages were created by the KEEP-Bench authors. |
| - Passages in the imbalanced setting are excerpts from publicly accessible web pages. Copyright remains with the original authors. These excerpts are provided solely for non-commercial research on context compression and may not be redistributed separately. |
|
|
| If you are a rights holder and want content removed, please open a discussion in the Community tab of this repository, and we will remove it promptly. |
|
|
| ## Citation |
|
|
| <!-- TODO: replace with the final reference --> |
| ```bibtex |
| @article{keepbench2026, |
| title = {{KEEP-Bench}: Evaluating Evidence Preservation in Context Compression}, |
| author = {TBD}, |
| journal = {arXiv preprint arXiv:TBD}, |
| year = {2026} |
| } |
| ``` |
|
|
