Updated Croissant file generation, citation file and provenance wording.

.pytest_cache/.gitignore

@@ -0,0 +1,2 @@
+# Created by pytest automatically.
+*

.pytest_cache/CACHEDIR.TAG

@@ -0,0 +1,4 @@
+Signature: 8a477f597d28d172789f06886806bc55
+# This file is a cache directory tag created by pytest.
+# For information about cache directory tags, see:
+#	https://bford.info/cachedir/spec.html

.pytest_cache/README.md

@@ -0,0 +1,8 @@
+# pytest cache directory #
+
+This directory contains data from the pytest's cache plugin,
+which provides the `--lf` and `--ff` options, as well as the `cache` fixture.
+
+**Do not** commit this to version control.
+
+See [the docs](https://docs.pytest.org/en/stable/how-to/cache.html) for more information.

.pytest_cache/v/cache/lastfailed

@@ -0,0 +1 @@
+{}

.pytest_cache/v/cache/nodeids

@@ -0,0 +1,16 @@
+[
+  "tests/test_attribution.py::AttributionTests::test_code_license_and_source_notices",
+  "tests/test_attribution.py::AttributionTests::test_scientific_author_order_and_software_author",
+  "tests/test_croissant.py::test_author_order_matches_the_paper_everywhere",
+  "tests/test_croissant.py::test_croissant_checksums_match_the_files_on_disk",
+  "tests/test_croissant.py::test_croissant_descriptor_matches_generator",
+  "tests/test_croissant.py::test_croissant_documents_the_release_layout",
+  "tests/test_croissant.py::test_croissant_optional_mlcommons_validator_passes",
+  "tests/test_croissant.py::test_croissant_prose_agrees_with_the_released_tables",
+  "tests/test_release.py::ReleaseTests::test_adapter_preserves_backbone_and_masks_redesign",
+  "tests/test_release.py::ReleaseTests::test_corrected_contig_preserves_motif_and_target",
+  "tests/test_release.py::ReleaseTests::test_reconstruction_rejects_unpinned_prosite",
+  "tests/test_release.py::ReleaseTests::test_reference",
+  "tests/test_release.py::ReleaseTests::test_upstream_motif_info_and_segment_checker",
+  "tests/test_release.py::ReleaseTests::test_upstream_summary_is_ordered_before_scoring"
+]

CHANGELOG.md

@@ -1,5 +1,36 @@
 # Changelog
 
+## Unreleased — metadata revision (data files unchanged)
+
+- Rewrote `croissant.json` through `scripts/build_croissant.py`: it now declares
+  conformance to both `http://mlcommons.org/croissant/1.1` and the
+  Responsible-AI vocabulary `http://mlcommons.org/croissant/RAI/1.0`, and adds
+  the full `rai:*` block (collection, preprocessing, annotation, use cases,
+  biases, limitations, personal-information status, consent, social impact and
+  `rai:hasSyntheticData: false`), `prov:wasDerivedFrom` / `prov:wasGeneratedBy`
+  / `prov:wasAssociatedWith` provenance for the PROSITE 2021_04 and RCSB/wwPDB
+  sources and the construction pipeline, plus a `prov` namespace in `@context`.
+- Added the verified publication metadata: `datePublished` 2026-11-02 (the
+  upload date of the v1.0.0 release), the dataset DOI `10.57967/hf/11435` as
+  `identifier`, `sameAs` and `citeAs` (registered 2026-11-04, pinned to commit
+  `b0312e0433191ec425c67e2f9a620c2f614551c2`), `creator` alongside `author`,
+  `keywords`, `alternateName` and `citation`.
+- Set `isLiveDataset` to false, added `sha256` digests to the five CSV
+  `FileObject` distributions and described the repository tree as a
+  `cr:FileSet` (`git+https`), since a Git tree has no single checksum.
+  `mlcroissant validate --jsonld croissant.json` now reports no error.
+- Documented the historical multi-state threshold ambiguity (see
+  construction/README.md) in the RAI limitations instead of silently
+  normalizing it, and added the release composition facts that were missing
+  from the metadata (187 unique source depositions, of which 16 serve more than
+  one case; 194 X-ray, 18 NMR, 7 EM and 1 combined NMR/scattering case).
+- Extended `tests/test_croissant.py` to pin the new metadata, the descriptor's
+  agreement with `provenance/release-sha256.txt` and the bytes on disk, and an
+  optional `mlcroissant` load when the package is installed.
+- Added the dataset DOI, URL, version, release date and license to
+  `CITATION.cff`; recorded the new descriptor facts in `provenance/README.md`.
+- No released data file changed: `provenance/release-sha256.txt` is untouched.
+
 ## BILBY-PRO 1.0.0 — initial public release preparation
 
 - Corrected `1A22_PS01352_c1` to meet its existing target length of 75:

CITATION.cff

@@ -1,9 +1,16 @@
 cff-version: 1.2.0
 message: >-
-  Please cite the scientific publication below when using BILBY-PRO, as well
-  as PROSITE, MotifBench, and the source PDB structures.
+  Please cite the BILBY-PRO dataset release with the DOI below, and the
+  scientific publication listed under preferred-citation, as well as PROSITE,
+  MotifBench, and the source PDB structures.
 title: "BILBY-PRO: PROSITE-derived functional motif scaffolding benchmark"
 type: dataset
+doi: 10.57967/hf/11435
+url: "https://huggingface.co/datasets/HFMI-PROFOUND/BILBY-PRO"
+repository-code: "https://huggingface.co/datasets/HFMI-PROFOUND/BILBY-PRO"
+version: 1.0.0
+date-released: 2026-11-02
+license: CC-BY-4.0
 authors: &scientific_authors
   - family-names: de Bruyn
     given-names: Emile
@@ -19,14 +26,14 @@ authors: &scientific_authors
     given-names: Aleena
   - family-names: Upadhyay
     given-names: Utkarsh
+  - family-names: Bauer
+    given-names: Stefan
   - family-names: Schug
     given-names: Alexander
-  - family-names: Bauer
+  - family-names: Kesselheim
     given-names: Stefan
   - family-names: Yakimovich
     given-names: Artur
-  - family-names: Kesselheim
-    given-names: Stefan
   - family-names: Bazarova
     given-names: Alina
 preferred-citation:

README.md

@@ -29,8 +29,8 @@ the Helmholtz Foundation Model Initiative (HFMI) of the Helmholtz Association.
 Associated publication: *Static Recovery Is Not Dynamic Stability:
 Dynamics-Aware Benchmarking of Protein Motif Scaffolding*, by Emile de Bruyn,
 Moritz Schäffler, Jannik Schneider, Sebastian R. Schmidt, Chanwoong Hwang,
-Aleena Siji, Utkarsh Upadhyay, Alexander Schug, Stefan Bauer, Artur Yakimovich,
-Stefan Kesselheim and Alina Bazarova. See [CITATION.cff](CITATION.cff).
+Aleena Siji, Utkarsh Upadhyay, Stefan Bauer, Alexander Schug, Stefan Kesselheim,
+Artur Yakimovich and Alina Bazarova. See [CITATION.cff](CITATION.cff).
 The first three authors contributed equally.
 
 ## Quick start
@@ -46,8 +46,12 @@ python3 scripts/validate_release.py
 The validator checks case and pattern counts, table relationships, residue
 coordinates, contig feasibility and checksums, including regression against
 the historical submission. See [croissant.json](croissant.json) for
-machine-readable table fields, links between tables and structure collections.
-For example, after `pip install mlcroissant`, read a local case with:
+machine-readable table fields, links between tables and structure collections,
+and for the Responsible-AI and provenance metadata (Croissant 1.1 and
+`http://mlcommons.org/croissant/RAI/1.0`, `rai:*` and `prov:*` properties).
+It is regenerated by `python3 scripts/build_croissant.py` and passes
+`mlcroissant validate --jsonld croissant.json` without errors. For example,
+after `pip install mlcroissant`, read a local case with:
 
 ```python
 import mlcroissant as mlc
@@ -56,6 +60,34 @@ dataset = mlc.Dataset("croissant.json", mapping={"test_cases.csv": "test_cases.c
 print(next(iter(dataset.records(record_set="test_cases"))))
 ```
 
+## Release integrity and citation
+
+`provenance/release-sha256.txt` pins all 445 released data files with one
+SHA-256 per line. The published release is
+[commit `b0312e0433191ec425c67e2f9a620c2f614551c2`](https://huggingface.co/datasets/HFMI-PROFOUND/BILBY-PRO/tree/b0312e0433191ec425c67e2f9a620c2f614551c2),
+uploaded on 2026-11-02; `main` may receive later revisions, so cite the commit
+or the dataset DOI rather than the branch. The dataset DOI
+`10.57967/hf/11435` (DataCite via Hugging Face, state findable, registered
+2026-11-04) is a `Dataset` record whose `contentUrl` is exactly that commit
+tree, so the DOI resolves to this same snapshot rather than to a moving branch:
+<https://api.datacite.org/dois/10.57967/hf/11435>. The v1.0.0 data files and
+their digests are unchanged by that registration. This release supersedes the
+earlier anonymous dataset, which remains the historical artifact behind the
+paper.
+
+`provenance/huggingface-manifest.json` records every hosted
+`resolve/main/<path>` URL with its exact content type, size and LFS marker: 450
+entries, namely the two structure directories, the five CSV tables, the hosted
+`README.md`, `LICENSE` and `.gitattributes`. The 440 PDB files are LFS objects;
+the CSV and JSON files are plain Git objects, so Git LFS filter rules cannot be
+used as an integrity check for them. `provenance/published-sha256.txt` retains
+the historical digests.
+
+This metadata revision changes no hosted data file. `croissant.json`,
+`CITATION.cff`, `CHANGELOG.md`, `construction/`, `provenance/` and `scripts/`
+are part of this repository release and are not yet in the hosted tree recorded
+by the manifest; refresh that manifest after they are uploaded.
+
 ## Dataset contents
 
 | File | Contents |
@@ -190,6 +222,14 @@ generation, refolding, Foldseek database search and full GPU evaluation were
 
 ## Attribution and citation
 
+Please cite the dataset release itself as described in
+[CITATION.cff](CITATION.cff): *BILBY-PRO: PROSITE-derived functional motif
+scaffolding benchmark*, DOI [10.57967/hf/11435](https://doi.org/10.57967/hf/11435),
+v1.0.0, commit `b0312e0433191ec425c67e2f9a620c2f614551c2`, and the associated
+publication *Static Recovery Is Not Dynamic Stability: Dynamics-Aware
+Benchmarking of Protein Motif Scaffolding*. The machine-readable preferred
+citation is the `citeAs` field of [croissant.json](croissant.json).
+
 Benchmark case definitions and added annotations are [CC BY 4.0](LICENSE).
 Construction code and release scripts are
 [Apache-2.0](LICENSE-APACHE-2.0), copyright 2026 Emile de Bruyn. PDB structures

croissant.json

@@ -6,6 +6,7 @@
     "sc": "https://schema.org/",
     "dct": "http://purl.org/dc/terms/",
     "containedIn": "cr:containedIn",
+    "prov": "http://www.w3.org/ns/prov#",
     "rai": "http://mlcommons.org/croissant/RAI/",
     "data": {
       "@id": "cr:data",
@@ -48,14 +49,34 @@
     }
   },
   "@type": "sc:Dataset",
-  "conformsTo": "http://mlcommons.org/croissant/1.1",
+  "conformsTo": [
+    "http://mlcommons.org/croissant/1.1",
+    "http://mlcommons.org/croissant/RAI/1.0"
+  ],
   "name": "BILBY-PRO",
-  "description": "BILBY-PRO benchmarks functional motif scaffolding in protein design with 220 cases from 174 PROSITE pattern entries. PROSITE 2021_04 (29 September 2021), maintained by SIB at https://prosite.expasy.org/, provided pattern definitions and 3D PDB cross-references. Its licensing conditions are at https://prosite.expasy.org/prosite_license.html. Structures and PROSITE-derived patterns retain upstream terms (PROSITE 2021_04 specifies CC BY-NC-ND 4.0); CC BY 4.0 covers independently constructed benchmark definitions and added annotations. Construction software by Emile de Bruyn is Apache-2.0.",
+  "alternateName": "BILBY-PRO benchmark",
+  "description": "BILBY-PRO benchmarks functional motif scaffolding in protein design with 220 cases from 174 PROSITE pattern entries. PROSITE 2021_04 (29 September 2021), maintained by SIB at https://prosite.expasy.org/, provided pattern definitions and 3D PDB cross-references. Its licensing conditions are at https://prosite.expasy.org/prosite_license.html. Structures and PROSITE-derived patterns retain upstream terms (PROSITE 2021_04 specifies CC BY-NC-ND 4.0); CC BY 4.0 covers independently constructed benchmark definitions and added annotations. Construction software by Emile de Bruyn is Apache-2.0. Release 1.0.0 is a frozen, checksum-pinned snapshot of the peer-review benchmark (published 2026-11-02, DOI 10.57967/hf/11435), not a continuously updated stream: pin commit b0312e0433191ec425c67e2f9a620c2f614551c2 and verify it against provenance/release-sha256.txt, because the main branch may receive later revisions. Responsible-AI and provenance metadata (rai:* and prov:* properties, conforming to http://mlcommons.org/croissant/RAI/1.0) document collection, preprocessing, annotation, intended uses, biases and limitations.",
   "version": "1.0.0",
   "url": "https://huggingface.co/datasets/HFMI-PROFOUND/BILBY-PRO",
-  "isLiveDataset": true,
+  "@id": "https://huggingface.co/datasets/HFMI-PROFOUND/BILBY-PRO",
+  "identifier": "https://doi.org/10.57967/hf/11435",
+  "sameAs": [
+    "https://doi.org/10.57967/hf/11435"
+  ],
+  "keywords": [
+    "biology",
+    "protein design",
+    "protein structure",
+    "motif scaffolding",
+    "benchmark",
+    "PROSITE",
+    "PDB",
+    "functional motifs"
+  ],
+  "datePublished": "2026-11-02",
+  "isLiveDataset": false,
   "license": "https://creativecommons.org/licenses/by/4.0/",
-  "author": [
+  "creator": [
     {
       "@type": "sc:Person",
       "name": "Emile de Bruyn",
@@ -126,6 +147,16 @@
         }
       ]
     },
+    {
+      "@type": "sc:Person",
+      "name": "Stefan Bauer",
+      "affiliation": [
+        {
+          "@type": "sc:Organization",
+          "name": "Helmholtz Munich"
+        }
+      ]
+    },
     {
       "@type": "sc:Person",
       "name": "Alexander Schug",
@@ -138,11 +169,15 @@
     },
     {
       "@type": "sc:Person",
-      "name": "Stefan Bauer",
+      "name": "Stefan Kesselheim",
       "affiliation": [
         {
           "@type": "sc:Organization",
-          "name": "Helmholtz Munich"
+          "name": "Jülich Supercomputing Centre, Forschungszentrum Jülich"
+        },
+        {
+          "@type": "sc:Organization",
+          "name": "Department of Physics, University of Cologne"
         }
       ]
     },
@@ -168,6 +203,108 @@
         }
       ]
     },
+    {
+      "@type": "sc:Person",
+      "name": "Alina Bazarova",
+      "affiliation": [
+        {
+          "@type": "sc:Organization",
+          "name": "Jülich Supercomputing Centre, Forschungszentrum Jülich"
+        }
+      ]
+    }
+  ],
+  "author": [
+    {
+      "@type": "sc:Person",
+      "name": "Emile de Bruyn",
+      "affiliation": [
+        {
+          "@type": "sc:Organization",
+          "name": "Jülich Supercomputing Centre, Forschungszentrum Jülich"
+        }
+      ]
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Moritz Schäffler",
+      "affiliation": [
+        {
+          "@type": "sc:Organization",
+          "name": "Jülich Supercomputing Centre, Forschungszentrum Jülich"
+        }
+      ]
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Jannik Schneider",
+      "affiliation": [
+        {
+          "@type": "sc:Organization",
+          "name": "Jülich Supercomputing Centre, Forschungszentrum Jülich"
+        }
+      ]
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Sebastian R. Schmidt",
+      "affiliation": [
+        {
+          "@type": "sc:Organization",
+          "name": "Jülich Supercomputing Centre, Forschungszentrum Jülich"
+        }
+      ]
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Chanwoong Hwang",
+      "affiliation": [
+        {
+          "@type": "sc:Organization",
+          "name": "Center for Advanced Systems Understanding (CASUS), Görlitz, Germany"
+        }
+      ]
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Aleena Siji",
+      "affiliation": [
+        {
+          "@type": "sc:Organization",
+          "name": "Helmholtz Munich"
+        }
+      ]
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Utkarsh Upadhyay",
+      "affiliation": [
+        {
+          "@type": "sc:Organization",
+          "name": "Jülich Supercomputing Centre, Forschungszentrum Jülich"
+        }
+      ]
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Stefan Bauer",
+      "affiliation": [
+        {
+          "@type": "sc:Organization",
+          "name": "Helmholtz Munich"
+        }
+      ]
+    },
+    {
+      "@type": "sc:Person",
+      "name": "Alexander Schug",
+      "affiliation": [
+        {
+          "@type": "sc:Organization",
+          "name": "Scientific Computing Center, Karlsruhe Institute of Technology"
+        }
+      ]
+    },
     {
       "@type": "sc:Person",
       "name": "Stefan Kesselheim",
@@ -182,6 +319,28 @@
         }
       ]
     },
+    {
+      "@type": "sc:Person",
+      "name": "Artur Yakimovich",
+      "affiliation": [
+        {
+          "@type": "sc:Organization",
+          "name": "Center for Advanced Systems Understanding (CASUS), Görlitz, Germany"
+        },
+        {
+          "@type": "sc:Organization",
+          "name": "Helmholtz-Zentrum Dresden-Rossendorf"
+        },
+        {
+          "@type": "sc:Organization",
+          "name": "Institute of Computer Science, University of Wrocław"
+        },
+        {
+          "@type": "sc:Organization",
+          "name": "Cluster of Excellence Physics of Life, TU Dresden"
+        }
+      ]
+    },
     {
       "@type": "sc:Person",
       "name": "Alina Bazarova",
@@ -197,15 +356,152 @@
     "@type": "sc:Organization",
     "name": "HFMI-PROFOUND"
   },
-  "citeAs": "Emile de Bruyn, Moritz Schäffler, Jannik Schneider, Sebastian R. Schmidt, Chanwoong Hwang, Aleena Siji, Utkarsh Upadhyay, Alexander Schug, Stefan Bauer, Artur Yakimovich, Stefan Kesselheim, Alina Bazarova. Static Recovery Is Not Dynamic Stability: Dynamics-Aware Benchmarking of Protein Motif Scaffolding.",
+  "citeAs": "@dataset{debruyn_bilby_pro_2026,\n  author       = {de Bruyn, Emile and Schäffler, Moritz and Schneider, Jannik and Schmidt, Sebastian R. and Hwang, Chanwoong and Siji, Aleena and Upadhyay, Utkarsh and Bauer, Stefan and Schug, Alexander and Kesselheim, Stefan and Yakimovich, Artur and Bazarova, Alina},\n  title        = {{BILBY-PRO: PROSITE-derived functional motif scaffolding benchmark}},\n  year         = {2026},\n  month        = nov,\n  publisher    = {Hugging Face},\n  version      = {1.0.0},\n  doi          = {10.57967/hf/11435},\n  url          = {https://huggingface.co/datasets/HFMI-PROFOUND/BILBY-PRO},\n  note         = {Corrected release of the benchmark introduced in \"Static Recovery Is Not Dynamic Stability: Dynamics-Aware Benchmarking of Protein Motif Scaffolding\"}\n}",
+  "citation": "Static Recovery Is Not Dynamic Stability: Dynamics-Aware Benchmarking of Protein Motif Scaffolding -- the publication that introduces and evaluates this benchmark; see CITATION.cff for the author-preferred citation.",
+  "prov:wasDerivedFrom": [
+    {
+      "@id": "https://prosite.expasy.org/",
+      "name": "PROSITE 2021_04 (SIB, Swiss-Prot + PROSITE)",
+      "description": "Release 2021_04, dated 29 September 2021, of the curated protein pattern and profile database.  Supplies the verbatim motif_name, pattern, motif_entry (PS accessions) and pdoc (PDOC accessions) values of motif_cases_metadata_croissant.csv and the PDB cross-references that define the candidate structure pool.  Licensing: https://prosite.expasy.org/prosite_license.html"
+    },
+    {
+      "@id": "https://ftp.expasy.org/databases/prosite/old_releases/prosite2021_04.tar.bz2",
+      "name": "Archived PROSITE 2021_04 release used as pipeline input",
+      "description": "prosite.dat parsed by construction/motifbench_prosite, pinned by scripts/reconstruct.py and documented in provenance/README.md (SHA-256 00f9598c5ddcef7b9d414cd24b968ad8a746e6b808e8fa525787f0e739f2e9c7)."
+    },
+    {
+      "@id": "https://www.rcsb.org/",
+      "name": "RCSB Protein Data Bank experimental structure entries",
+      "description": "187 unique macromolecular depositions behind the 220 cases (164 X-ray diffraction, 15 solution NMR, 7 electron microscopy, 1 combined NMR and solution scattering; counted per case instead, the split is 194, 18, 7 and 1); all atomic coordinates and residue identities in motif_pdbs and original_pdbs.  Entries remain under their deposited PDB licenses: https://www.wwpdb.org/"
+    },
+    {
+      "@id": "https://github.com/blt2114/MotifBench",
+      "name": "MotifBench reference test-case table",
+      "description": "The historical MotifBench test_cases.csv used only to drop candidate cases that duplicate an existing MotifBench problem; the released file layout and group definitions follow it."
+    },
+    {
+      "@id": "https://huggingface.co/datasets/anonymous-motif-scaffolding/prosite_functional_motif_scaffolding_benchmark/tree/b16ffd61dc6b84fffa10f007282d5ac22ba16b95",
+      "name": "Peer-review snapshot of this benchmark",
+      "description": "The anonymous submission release from which this corrected v1.0.0 snapshot derives: the same 220 cases, the same 440 PDB files and the same motif metadata table, with one infeasible contig gap corrected and the cluster case identifiers renamed from the _vN to the _cN suffix (see CHANGELOG.md)."
+    }
+  ],
+  "prov:wasGeneratedBy": {
+    "@type": "prov:Activity",
+    "@id": "https://huggingface.co/datasets/HFMI-PROFOUND/BILBY-PRO/tree/main/construction",
+    "name": "BILBY-PRO construction pipeline and corrected release",
+    "description": "Collection window: PROSITE release 2021_04 (29 September 2021) and PDB entries retrieved from live RCSB servers during the historical construction run, whose exact date and environment were not logged; the corrected release was published on 2026-11-02.  Geographic scope: not applicable, the sources are global scientific repositories.  Protocol (construction/motifbench_prosite, run as python -m motifbench_prosite --cli construction/configs/reference.yaml download,extract,analyze,export,benchmark): parse prosite.dat and keep the PROSITE entries carrying at least one 3D PDB cross-reference, filtered to pattern spans of 12-100 residues and at most 25 PDB records per pattern; retrieve those entries with Biopython's PDBList, converting mmCIF above 10 MB; match each IUPAC pattern against polymer sequences and merge contiguous matches into motif segments, absorbing short internal wildcard stretches as redesignable positions and keeping longer ones as non-motif context; keep candidates with at least 5 structural instances whose median pairwise CA RMSD falls below the conservation threshold (min_instances 5, rmsd_threshold 2.0 in configs/reference.yaml); cluster motif fragments with Foldseek (min_seq_id 0.5, cov_mode 1, cov_threshold 0.8) to derive cluster ratios, group 4 multi-state labels and cluster_membership.csv; discard cases that duplicate the reference MotifBench problems; tag assembly context as core (179 cases), ribosomal (30) or huge_assembly (11); and export MotifBench-compatible motif-only PDB files, source PDB files, segment and redesignable-residue annotations and contigs over the reference scaffold lengths 75, 100, 125, 150, 175, 200 and 225, with coordinates and residue labels normalized on export.  Release 1.0.0 additionally corrected the 1A22_PS01352_c1 contig gap (25-25 to 21-25), standardized the 400 cluster membership identifiers to the _cN suffix and pinned SHA-256 digests of all 445 released data files in provenance/release-sha256.txt.",
+    "prov:wasAssociatedWith": [
+      {
+        "@type": "prov:SoftwareAgent",
+        "@id": "https://huggingface.co/datasets/HFMI-PROFOUND/BILBY-PRO/tree/main/construction/motifbench_prosite",
+        "name": "motifbench_prosite (Apache-2.0)"
+      },
+      {
+        "@type": "prov:SoftwareAgent",
+        "@id": "https://github.com/steineggerlab/foldseek",
+        "name": "Foldseek (structural clustering; version not recorded)"
+      },
+      {
+        "@type": "prov:SoftwareAgent",
+        "@id": "https://biopython.org/",
+        "name": "Biopython (PDBList, PDBParser, MMCIFParser, PDBIO)"
+      }
+    ]
+  },
+  "rai:dataCollection": "No human subjects, crowdsourcing or private data were involved: the benchmark was assembled programmatically from public, curated scientific resources by the construction/motifbench_prosite pipeline, whose stages are (1) parse PROSITE release 2021_04 prosite.dat, (2) retrieve the 3D-cross-referenced PDB entries with Biopython's PDBList, (3) match IUPAC patterns to polymer sequences and extract residue-level coordinates, (4) compute structural conservation and Foldseek clustering statistics, (5) filter, normalize and export MotifBench-compatible files.  All 220 cases derive from experimental depositions: 194 from X-ray diffraction, 18 from solution NMR, 7 from electron microscopy and 1 from combined NMR and solution scattering.  The collection is unrestricted with respect to geography, population or organism; it is bounded only by which protein families PROSITE 2021_04 patterns and the PDB had solved by that release.  The source vocabulary and the PDB cross-references are therefore fixed at 2021-09-29, while the underlying depositions keep their own deposition dates and span several decades of structural biology; the construction run itself was not logged, so its downloads reflected the live RCSB servers of that moment rather than a pinned snapshot.  This corrected release was published on 2026-11-02 and its dataset DOI was registered on 2026-11-04.",
+  "rai:dataCollectionType": [
+    "Secondary Data analysis",
+    "Software Collection"
+  ],
+  "rai:dataCollectionRawData": "Three inputs, none of them produced for this dataset: PROSITE release 2021_04 prosite.dat of 29 September 2021 (SHA-256 00f9598c5ddcef7b9d414cd24b968ad8a746e6b808e8fa525787f0e739f2e9c7, tarball SHA-256 32db1518ae124f42f659b4a1151e4510b3aae6cb6a87f379508783342a2c99d7, documented in provenance/README.md and pinned by scripts/reconstruct.py), which provides the verbatim pattern, motif_name, motif_entry and pdoc strings as well as the PDB cross-references; the experimentally determined PDB entries referenced by those patterns, 187 unique depositions that supply every coordinate distributed here; and the historical MotifBench reference table, used only to remove duplicated cases.  No per-case manual measurement, wet-lab experiment or human annotation produced any value in the release.",
+  "rai:dataCollectionMissingData": "A case exists only if its PROSITE entry carries a 3D cross-reference in the 2021_04 release, matches a downloadable structure of at most 10 MB, yields at least 5 structural instances and passes the median pairwise RMSD screen; entries failing any of those steps, patterns whose matches were skipped after download failures, and functions without a PROSITE signature or without an experimental structure are absent.  Because the historical run log, PDB snapshot and Foldseek version were not retained, the rejected and skipped inventories cannot be enumerated.  Within the released cases, water atoms were stripped and ligand, ion and nucleic-acid coordinates are excluded from motif_pdbs, which holds polymer atoms only; original_pdbs retain the non-polymer entities of the deposition.  Multi-chain and interface sites are therefore represented by a single chain only.",
+  "rai:dataCollectionTimeFrame": "2021-09-29",
+  "rai:dataImputationProtocol": "No values were imputed, interpolated or modelled.  Candidates that were unresolved, too large, failed to download or lacked 5 conserved instances were dropped rather than filled in, and no case carries placeholder coordinates.  The only post-hoc correction is documented in CHANGELOG.md: at release 1.0.0 the contig of case 1A22_PS01352_c1 shortened an internal gap minimum from 25 to 21 residues so that the specified contig can reach the 75-residue target length, leaving motif residues, order, maximum gap and target length unchanged.",
+  "rai:dataPreprocessingProtocol": [
+    "Pattern span filter 12-100 residues and at most 25 PDB records per pattern (prosite_filters in configs/reference.yaml).",
+    "Contiguous pattern matches merged into motif segments; short internal wildcards absorbed as redesignable motif positions, longer wildcard spans kept as non-motif context, which is why 207 cases hold a single motif segment, 11 hold two and 2 hold three (235 segments in total).",
+    "Structural conservation filter: minimum 5 instances and median pairwise CA RMSD below the configured 2.0 A threshold.",
+    "Foldseek clustering of motif fragments (min_seq_id 0.5, cov_mode 1, cov_threshold 0.8) for cluster ratios, multi-state group 4 labels and the 400-row cluster_membership.csv.  Note that configs/reference.yaml records multi_state_ratio 0.75 and multi_state_min_n 3, but PipelineConfig does not load those keys and the export code defaults to 0.9 and 10; the historical settings are unknown, so the 96 group 4 labels are reported as released rather than re-derived.",
+    "Removal of cases duplicating the reference MotifBench problems.",
+    "Assembly context tagging from deposition metadata: pdb_n_chains, pdb_total_len, pdb_max_chain_len, pdb_has_nucleic_acid, pdb_is_ribosomal and pdb_is_huge_assembly, mapped to core, ribosomal or huge_assembly.",
+    "Normalization on export: residue renumbering, chain remapping recorded in normalisation_chain_map, motif-only PDB with one chain per segment, water atoms stripped and non-polymer entities excluded.",
+    "Contig construction against the reference scaffold lengths 75, 100, 125, 150, 175, 200 and 225, with overly large internal gap minima reduced only as far as needed to fit the target length."
+  ],
+  "rai:dataPreprocessingPlatform": [
+    "Python 3 command-line pipeline in construction/motifbench_prosite plus the external Foldseek binary, run from this repository; no hosted data-processing service or managed cluster is documented."
+  ],
+  "rai:preprocessingSoftwarePackages": [
+    "Dependencies declared by construction/pyproject.toml: python>=3.10, pandas>=1.5, biopython>=1.81, numpy, xarray, scipy, MDAnalysis, PyYAML and tqdm, plus Foldseek for structural clustering. The exact versions of the historical construction run are not established (construction/README.md), so they are not asserted here."
+  ],
+  "rai:preprocessingVerification": [
+    "python3 -m pytest tests -q runs the release suite: Croissant descriptor reproducibility, table and identifier integrity, attribution and checksum tests.",
+    "python3 scripts/validate_release.py re-derives every released residue set from the distributed PDB files, verifies that each motif PDB contains exactly the annotated residues with one CA atom each, checks that all exported CA coordinates match the corresponding source residues, confirms all 220 contigs are feasible for their target lengths, and compares 445 pinned digests in provenance/release-sha256.txt with the files and with the Hugging Face manifest.",
+    "python3 -m mlcroissant validate --jsonld croissant.json (mlcroissant 1.1.1) parses the descriptor and its record sets.",
+    "A fresh reconstruction of the published selection has not been reproduced and cannot be, so the filters above are documented rather than re-executed."
+  ],
+  "rai:dataAnnotationProtocol": [
+    "Motif boundaries come from PROSITE pattern matches mapped onto residue indices and coordinates of the source chain; redesignable_segments records the matched positions whose identity may be redesigned and not_motif_segments is empty in this release.",
+    "Sequence annotations motif_entry, motif_name, pattern, pdoc and pdoc_url are copied verbatim from prosite.dat of release 2021_04, so they carry the curation state of PROSITE rather of this benchmark.",
+    "Structural annotations median_pairwise_rmsd, rmsd_instance_count and rmsd_common_length are computed by MDAnalysis over the available instances of the motif segment set (5 to 316 instances, common aligned length 12 to 78 residues here).",
+    "multi_state and group 4 come from Foldseek cluster ratios; assembly flags come from deposition metadata as described above.",
+    "Contig annotations (contigs, contig, contig_original, length) are generated from the segment geometry by the export stage.",
+    "No human annotation, no annotation guideline document and no annotator-agreement measurement exist for these fields; quality control is programmatic (see rai:preprocessingVerification).",
+    "Motif function is not annotated here: it is inherited from the PROSITE signature and its PDOC page, reachable through pdoc_url."
+  ],
+  "rai:dataAnnotationPlatform": [
+    "Not applicable: no human annotation platform, labeling service or crowdsourcing system was used; all attributes are derived by the construction pipeline and stored in motif_cases_metadata_croissant.csv, contigs.csv, contig_specifications.csv and cluster_membership.csv."
+  ],
+  "rai:annotationsPerItem": [
+    "One automated annotation record per case: 34 columns in motif_cases_metadata_croissant.csv, 9 in test_cases.csv, plus one contig row each; 0 human annotations per item.  cluster_membership.csv adds 180 non-representative member rows, so 400 member annotations cover the 341 clustered PROSITE entries behind the 220 released representatives."
+  ],
+  "rai:machineAnnotationTools": [
+    "MDAnalysis (pairwise CA RMSD), Foldseek (structural clustering and multi-state labelling), Biopython (structure parsing, PDB retrieval, fragment writing), pandas and numpy (table construction)."
+  ],
+  "rai:annotatorDemographics": [
+    "Not applicable: no human annotators, and therefore no inter-annotator agreement, were involved in producing any released annotation."
+  ],
+  "rai:dataUseCases": [
+    "Intended use: standardized evaluation of motif-scaffolding generative models, comparing whether a model can build a scaffold of the requested length around a fixed functional motif, with target coverage and structural recovery computed by the companion evaluation specification and per-solution metrics.",
+    "Intended use: stratified analysis by assembly context (core, ribosomal, huge_assembly), by MotifBench group (1: 113, 2: 9, 3: 2, 4 multi-state: 96 cases), by motif length (11 to 53 residues) and by structural conservation (median_pairwise_rmsd, rmsd_instance_count).",
+    "Intended use: a documented, checksum-pinned and PROSITE-annotated benchmark panel for reporting and reproducing protein design results, and a starting point for dynamics-aware evaluation that requires separate simulation, since no trajectories are included here.",
+    "Not intended: as a source of experimentally verified function, as a biological safety or risk assessment, as a therapeutic or engineering specification, or as a general-purpose training set; it is a small evaluation panel with no train, validation or test split.",
+    "Not intended: as an unbiased measure of generalization, because overlap with the training data of evaluated models cannot be excluded and because the case set is heavily biased toward single conformations and contiguous motifs."
+  ],
+  "rai:dataBiases": [
+    "Source-annotation bias: coverage follows PROSITE 2021_04 signatures and the historically well-studied families that have many solved structures, so abundant superfamilies are over-represented and under-solved protein classes are absent altogether.",
+    "Structural-method bias: 194 of 220 cases come from X-ray diffraction, only 18 from solution NMR, 7 from electron microscopy and 1 from solution scattering, favouring rigid, crystallizable, soluble domains; membrane and transient complexes are under-represented.",
+    "Selection bias toward rigidity: the 5-instance and RMSD-conservation filters prefer motifs that recur in nearly identical conformations, so flexible, context-dependent and induced-fit sites are systematically under-represented, and the scarcity of group 2 (9) and group 3 (2) cases reflects the contiguous, near-contiguous nature of PROSITE patterns rather than the true prevalence of multi-segment sites.",
+    "Redundancy bias: 220 representatives stand for 341 clustered PROSITE entries and 16 source depositions serve more than one case (one deposition serves 19), so cases are not independent samples and aggregated scores are dominated by the largest clusters.",
+    "Conformational bias: each case pins one representative state, so even the 96 multi-state cases describe a distribution only through cluster ratios, not through the alternative coordinates themselves.",
+    "Train-test contamination: the sources are public and widely used, so pretrained structure models may have been trained on these or closely related structures; no held-out guarantee, sequence-identity split or date cutoff is provided, and the historical construction PDB snapshot was not recorded.",
+    "Taxonomic and functional bias toward model organisms and enzymatic families represented in PROSITE; the 30 ribosomal and 11 huge_assembly cases are single-chain excerpts of larger complexes and do not carry their partners."
+  ],
+  "rai:dataLimitations": [
+    "The benchmark covers only structurally conserved, contiguous or near-contiguous within-chain segments from PROSITE-linked experimental structures; it under-represents discontinuous functional sites formed by distant sequence positions, motifs with no solved structure and functions that depend on ligands, ions, nucleic acids, membranes or interaction partners, which are excluded from the motif files.",
+    "Only one state per case is released, so a benchmark score does not characterise conformational ensembles, folding pathways or thermodynamic stability, and the paper's central finding is precisely that static recovery and dynamic stability disagree.",
+    "High static success on these cases establishes neither biochemical activity of the transferred motif nor expression, solubility or safety of the design; nothing here substitutes for experimental testing.",
+    "Statistical caution: 220 cases from 174 PROSITE entries and 187 depositions are not independent observations, no train/validation/test partition or cluster-disjoint split is provided, group 2 and group 3 contain only 9 and 2 cases, and no per-case confidence or uncertainty estimate accompanies the annotation values.",
+    "Reproducibility limits: the historical run log, source revision, PDB snapshot, locked environment and Foldseek version were not retained and no reconstruction has been shown to reproduce the published selection, so the released files are the authoritative definition; 9 cases report median pairwise RMSD below 0.1 A and 5 cases above the 2.0 A threshold recorded in configs/reference.yaml, which is consistent with the documented uncertainty about the historical filter settings and makes strict threshold-based reproduction impossible.",
+    "No functional annotation beyond the inherited PROSITE signature, no experimental activity data, no dynamics information, no designed solutions and no evaluation trajectories are included; sequence-side designs must be produced by the user's model and judged by the companion evaluation code."
+  ],
+  "rai:personalSensitiveInformation": [
+    "The dataset contains no personal, human-subject or user-generated data: the released files carry protein sequence patterns, residue numbering, atomic coordinates and structural statistics only.  No names, contact details, health information, identifiers of individuals, positions or other sensitive attributes are distributed; the demographic axes used for people-centred datasets (gender, ethnicity, socio-economic status, geography, language, age, culture, religion, disability status) do not apply, and species or organism metadata is limited to what the PDB deposition names in structure titles.  Author and affiliation metadata describes the researchers who produced the benchmark, not data subjects, and no crowdsourced or third-party labour contributed to it."
+  ],
+  "rai:dataSocialImpact": "Positive: it makes a class of protein design capability measurable and reproducible against documented PROSITE and PDB sources, which supports transparent comparison, failure analysis and, in the long run, more reliable computational design of enzymes and binders.  Negative and dual-use: the same capability is shared with any user, and progress on static motif transfer can be over-interpreted as progress toward functional proteins; benchmark results can also be biased by the structural coverage of the source databases, and inflated claims about designed proteins could misdirect downstream research.  Mitigations: the panel ships evaluation specifications only, with no model weights, no generation code and no wet-lab protocols; the documented representativeness and reproduction limits above constrain the claims that can be made from a score; experimental validation and applicable biosafety review remain the responsibility of downstream users.",
+  "rai:dataConsent": "Not applicable: no human participants or personal data were collected, so no consent process was involved.  Upstream consent and release terms apply to the redistributed PDB and PROSITE content: PROSITE 2021_04 specifies CC BY-NC-ND 4.0 for its pattern definitions and documentation and PDB entries remain under their deposited licenses.",
+  "rai:hasSyntheticData": false,
   "distribution": [
     {
-      "@type": "cr:FileObject",
+      "@type": "cr:FileSet",
       "@id": "repository",
       "name": "repository",
-      "description": "Hugging Face Git repository; main can change after release.",
+      "description": "Released tree of the Hugging Face Git repository: the 220 motif and 220 original coordinate files of the two structure FileSets below, plus the CSV tables and the provenance, reconstruction and validation material.  Frozen at commit b0312e0433191ec425c67e2f9a620c2f614551c2; main may receive later revisions, so cite the commit or the DOI rather than the branch.",
       "contentUrl": "https://huggingface.co/datasets/HFMI-PROFOUND/BILBY-PRO",
-      "encodingFormat": "git+https"
+      "encodingFormat": "git+https",
+      "includes": [
+        "motif_pdbs/*.pdb",
+        "original_pdbs/*.pdb"
+      ]
     },
     {
       "@type": "cr:FileObject",
@@ -482,7 +778,7 @@
           "@type": "cr:Field",
           "@id": "motif_cases_metadata_croissant/motif_name",
           "name": "motif_name",
-          "description": "PROSITE name of the selected pattern entry.",
+          "description": "PROSITE entry name, copied verbatim from the PROSITE 2021_04 release; upstream PROSITE terms apply to this string.",
           "dataType": "sc:Text",
           "source": {
             "fileObject": {
@@ -497,7 +793,7 @@
           "@type": "cr:Field",
           "@id": "motif_cases_metadata_croissant/pattern",
           "name": "pattern",
-          "description": "PROSITE source-derived sequence pattern string.",
+          "description": "PROSITE sequence pattern, copied verbatim from the PROSITE 2021_04 release in PROSITE notation (CYS and HIS positions and position counts); the matching segment is fixed during scaffolding and its residue identities must be recovered, while redesignable_segments lists the positions whose amino acid identity may change.  Upstream PROSITE terms apply to this string.",
           "dataType": "sc:Text",
           "source": {
             "fileObject": {