Published November 5, 2026 | Version v2

PQID v2: A License-Filtered and Process-Audited Quantum-Programming Instruction Dataset

Authors/Creators

  • 1. Yonsei University

Description

PQID v2 is a license-filtered and process-audited quantum-programming instruction dataset. Outputs are predominantly generated diagnostic or explanatory teacher text, alongside retained source-code targets and one curated-repair lineage.

The October 29 revision provides two overlapping profiles: assessment_primary, 428,874 records (71,479 seeds and 357,395 paraphrases), and declared_permissive, 368,376 records (61,396 seeds and 306,980 paraphrases). Five paraphrases reuse each seed answer. The 550,314-record construction object is restricted internal material and is not supplied here.

Primary targets comprise 414,774 teacher-text, 14,094 retained source-code and six curated-repair records; companion targets comprise 359,364 teacher-text, 9,006 retained source-code and the same six repairs. Each ZIP contains three compressed JSONL splits, a standard-library reader, manifests/checksums, schema, source-linked public decision/notice derivatives and applicable notices. Extract each profile into its own directory. The primary archive reads assessment_primary by default; the companion archive reads declared_permissive. Profile choice is explicit in the reader.

The incorporated revision records 213 confirmed owner recovery cases and exact component-scope decisions; this is not blanket clearance of every associated source. Reciprocal licences are distinguished from permissive licences. The primary retains existing restricted and unresolved source categories; its inclusion policy is not unrestricted redistribution or model-training permission. The companion follows the narrower adopted source policy while retaining notice and attribution duties. CC BY 4.0 applies to PQID-authored documentation, metadata design, audit summaries and database arrangement, not as a replacement for upstream component terms. Source licences do not independently establish rights in generated answers. Credit and custom obligations, including Ryan Thompson's covered RandomNumber.py contribution, are retained.

A separate preregistered study selected 1,450 cases from an earlier frozen 422,580-record frame; fourteen external contributors completed 1,067. The later v2 additions were not thereby independently assessed. Source execution, prompt similarity, assisted ratings and external answer judgments have different evidential scopes. Most targets remain unassessed; the standard splits are not repository-disjoint. Source context may require a separately permitted, exact version; no complete source cache is included.

Private correspondence, participant records, internal reviewer documents and manuscript Markdown drafts are excluded. Public registries are declared derivatives of the private evidence originals; REGISTRY_RESOLUTION.json resolves their IDs without claiming byte identity. The accompanying reader and deterministic export checks do not constitute a full public reconstruction of the original API-generated corpus. The historical v1.0.2 record remains unchanged.

Files

PQID-v2-assessment_primary.zip

Files (1.5 GB)

Name Size
md5:0c13b08e50a2566f1174f9e369c38364
827.2 MB Preview Download
md5:4a4f675afe056c438a0413e193951969
651.8 MB Preview Download
md5:9f9c5e7671d9070aa574d7474c50b78c
1.4 kB Preview Download
md5:ab428510a5b28d4895321fe1b0c1cf1d
290 Bytes Preview Download