Add files using upload-large-folder tool

CHECKSUMS.sha256

@@ -0,0 +1 @@
+a6c68146cfe13ed93b14486ff16020ea1af29e44049a3e6be00e31652d0bb2c0  data/train-00000-of-00001.parquet

LICENSE

@@ -0,0 +1,14 @@
+BASIC 10K is made available under the Open Data Commons Attribution
+License, version 1.0 (ODC-By-1.0):
+
+https://opendatacommons.org/licenses/by/1-0/
+
+The dataset is derived from the Allen Institute for AI's Dolci-Think-SFT-7B
+dataset. Redistributions and derivative works must preserve attribution to that
+source and comply with its terms and responsible-use guidance:
+
+https://huggingface.co/datasets/allenai/Dolci-Think-SFT-7B
+https://allenai.org/responsible-use
+
+Copyright and database rights in individual source records remain with their
+respective owners where applicable.

README.md

@@ -1,3 +1,165 @@
 ---
-license: mit
+pretty_name: BASIC 10K
+license: odc-by
+language:
+  - en
+  - multilingual
+task_categories:
+  - text-generation
+size_categories:
+  - 10K<n<100K
+tags:
+  - chain-of-thought
+  - reasoning
+  - supervised-fine-tuning
+  - interpretability
+  - controllability
+configs:
+  - config_name: default
+    data_files:
+      - split: train
+        path: data/train-*
 ---
+
+# BASIC 10K
+
+**BASIC** (Better Abstractions with Structure for Interpretability and Control)
+is a supervised fine-tuning dataset for structured long-form reasoning. It contains
+10,000 teacher-generated traces that express reasoning as alternating short decisions
+and longer executions:
+
+```text
+<think>
+<steer>Choose the next high-level step</steer>
+<exec>Carry out that step in detail...</exec>
+...
+</think>
+Final answer
+```
+
+The dataset accompanies the paper *BASIC: Isolating Key Decisions During Long LLM
+CoT*. It is intended to make high-impact decisions in long reasoning traces easier to
+inspect, intervene on, and sample independently from lower-level execution text.
+
+## Dataset construction
+
+Prompts were stratified from the
+[Dolci-Think-SFT-7B](https://huggingface.co/datasets/allenai/Dolci-Think-SFT-7B)
+dataset across science, Python code, mathematics, multilingual chat, and
+instruction-following. A BASIC teacher generated a complete structured response for
+each selected prompt. The released records are the resulting distillation examples.
+
+The teacher was trained to produce the alternating format and to make each
+`<exec>` span defer to the immediately preceding `<steer>` decision, including for
+strongly off-policy steering instructions. This addresses an early failure mode in
+which models reproduced the markup but did not reliably follow the hierarchy.
+
+The release contains 10,000 traces drawn from 2,138 source prompt indices. Multiple
+rollouts may therefore share a source prompt. Traces are capped at 16,384 tokens.
+
+## Dataset structure
+
+The repository contains one `train` split in Parquet format. Each record has these
+fields:
+
+| Field | Type | Description |
+| --- | --- | --- |
+| `source_index` | integer | Index of the selected source prompt. |
+| `custom_id` | string | Identifier inherited from the source example. |
+| `ground_truth` | string | Source-provided reference answer when available; otherwise empty. |
+| `prompt` | list of messages | System/user conversation before the generated assistant response. |
+| `messages` | list of messages | Complete training conversation, including the structured assistant response. |
+| `completion` | string | Generated assistant response containing alternating `<steer>` and `<exec>` spans. |
+| `final_answer` | string | Answer text extracted from the generated completion. |
+| `steer_count` | integer | Number of `<steer>` spans in the completion. |
+| `exec_count` | integer | Number of `<exec>` spans in the completion. |
+| `length_tokens_total` | integer | Token count recorded during dataset generation. |
+| `rollout_index` | integer | Rollout number for the source prompt. |
+
+Every record is non-null. The `messages` field is the most direct input for chat-style
+supervised fine-tuning; `prompt` and `completion` support completion-style pipelines.
+
+## BASIC format
+
+A valid BASIC trace contains decision--execution pairs
+
+```text
+(<steer> decision </steer>, <exec> execution </exec>)
+```
+
+where the decision is a short, imperative natural-language instruction and the
+execution carries it out in detail. In the paper's decoding setup, `<steer>` spans are
+limited to 15 tokens and `<exec>` spans to 512 tokens. Structured decoding enforces the
+grammar and permits separate sampling parameters for the two span types.
+
+Training requires adding these four tokens to the tokenizer:
+
+```text
+<steer>  </steer>  <exec>  </exec>
+```
+
+## Paper training setup
+
+The experiments used a deterministic 95/5 training/evaluation split (9,500/500
+records; shuffle seed 42), assistant-only supervision, packed 16,384-token sequences,
+and a checkpoint after two epochs. The reported optimizer settings were
+`adamw_torch`, learning rate `7e-6`, betas `(0.9, 0.95)`, epsilon `1e-6`, weight decay
+`0.1`, and bfloat16 precision.
+
+These settings document the paper's use of the data; they are not requirements for
+other models or training stacks.
+
+## Intended use
+
+This dataset is intended for research on:
+
+- supervised fine-tuning of structured reasoning models;
+- inspection and intervention at explicit reasoning boundaries;
+- faithfulness to on-policy and off-policy steering decisions;
+- decision-level exploration and adaptive inference; and
+- interpretability and control of long-form reasoning.
+
+It is not intended as a source of verified factual answers or as a deployment-ready
+safety mechanism. Generated traces and final answers may be incorrect.
+
+## Limitations and responsible use
+
+The accompanying study evaluates mathematical reasoning, two model sizes, and
+controlled diversity tasks. Generalization to other domains, larger scales, longer
+trajectories, and tool use has not been established.
+
+Stronger control over chain-of-thought can support auditing, but it can also enable
+harmful steering or create a misleading appearance of faithfulness. The paper
+recommends access controls, logging, isolation from untrusted inputs, and independent
+outcome checks before real-world use. The source data is broad and multilingual, and
+generated content can inherit errors, biases, unsafe material, or personal information
+from its prompts and teacher model. Users should perform filtering appropriate to their
+application.
+
+## License and attribution
+
+This derived dataset is released under the
+[Open Data Commons Attribution License 1.0](https://opendatacommons.org/licenses/by/1-0/)
+(ODC-By-1.0). It is derived from the Allen Institute for AI's
+[Dolci-Think-SFT-7B](https://huggingface.co/datasets/allenai/Dolci-Think-SFT-7B),
+which is also distributed under ODC-By-1.0 and intended for research and educational
+use under [Ai2's Responsible Use Guidelines](https://allenai.org/responsible-use).
+Downstream users must preserve attribution and review the upstream terms.
+
+## Citation
+
+The paper is under anonymous review. Complete bibliographic information will be added
+when the review period ends. During review, cite the accompanying paper by title:
+
+```bibtex
+@misc{anonymous_basic_2026,
+  title = {{BASIC}: Isolating Key Decisions During Long {LLM} {CoT}},
+  author = {{Anonymous Authors}},
+  year = {2026}
+}
+```
+
+## Integrity
+
+The SHA-256 checksum of the released Parquet shard is recorded in
+`CHECKSUMS.sha256`.

data/train-00000-of-00001.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a6c68146cfe13ed93b14486ff16020ea1af29e44049a3e6be00e31652d0bb2c0
+size 92321967