Jack-Jieke-Wu/Paper-Writing-Exam-Source-Archive
Paper-Writing Exam Source Archive This dataset is the immutable provenance archive for Paper-Writing-Exam. It stores the task-paper registry and the original workflow inputs used to construct a fixed task release. It is an archive product, not a runnable Harbor task dataset. The initial immutable archive release is v0.1.0, target commit 328f2ca7ee8b68cbf2b5af60ad9ff1b85f47cce0. Dataset relationship Dataset Purpose May Harbor run it? Paper-Writing-Exam… See the full description on the dataset page: https://huggingface.co/datasets/Jack-Jieke-Wu/Paper-Writing-Exam-Source-Archive.
Paper-Writing Exam Source Archive
This dataset is the immutable provenance archive for Paper-Writing-Exam. It stores the task-paper registry and the original workflow inputs used to construct a fixed task release. It is an archive product, not a runnable Harbor task dataset.
The initial immutable archive release is `v0.1.0`, target commit 328f2ca7ee8b68cbf2b5af60ad9ff1b85f47cce0.
Dataset relationship
No task directory, solution, verifier, or trial artifact is part of this archive. Archive updates do not alter a task release or its checksum.
Current coverage
The initial archive maps the 274-task Paper-Writing-Exam release at v0.4.1 / 367bbf67ea05d0ec3d580e062aadf5636b45fc65:
- 51
paperwrite-bench-shorttasks; - 200
paperwritingbench-sparse-plotofftasks; - 22
lifesci-paperrecon-shorttasks; and - one first-party
hello-worldsmoke task, which correctly has no source paper or external archive input.
It contains 273 archived upstream source trees and a per-file SHA-256 manifest. The LifeSci source records include the pinned arXiv identifier/version, source URL, paper license, code locator, and code revision recorded by the build. The two adapted upstream benchmark families use their official stable sample IDs and pinned upstream revisions where no canonical arXiv mapping was available in the upstream dataset.
Files
README.md
archive-metadata.json
registry/task-paper-registry.jsonl
manifests/source-archive-manifest.jsonl
sources/registry/task-paper-registry.jsonl has one record for each (config, task_id, task release). It links the task identity and checksum to the upstream sample or paper, fixed conversion revision, and source archive tree. The first-party smoke task is explicitly marked not_applicable rather than being assigned a fictional paper.
manifests/source-archive-manifest.jsonl records every archived file's repository-relative path, size, SHA-256, source kind, and paper/source metadata.
How to use this archive
Use an immutable archive revision when inspecting provenance:
hf download Jack-Jieke-Wu/Paper-Writing-Exam-Source-Archive \
registry/task-paper-registry.jsonl \
--type dataset \
--revision <archive-tag-or-commit> \
--local-dir ./paper-writing-exam-source-archiveUse the registry to locate a source tree and compare it with a task's release identity. Do not mount, copy, or expose these files as a Harbor task's environment/materials/, solution/, or tests/ content. The task dataset remains self-contained and is the sole source for evaluation input.
Licensing and redistribution
Each archive record preserves its upstream locator and license metadata. The archive does not replace the original licenses of PaperWrite-Bench, PaperWritingBench/PaperOrchestra, LifeSci source papers, code repositories, or conference templates. Reuse and redistribution must comply with the per-source terms. A code locator with no declared license remains a locator; it is not a new permission grant.
