---
license: apache-2.0
task_categories:
  - text-classification
language:
  - en
tags:
  - mock-data
  - semantic-classification
  - enterprise-schemas
  - odata
---

# MockGen Classifier Data v1

## Dataset Summary

This dataset contains publication-safe schema-property rows used to train, validate, and
analyze the MockGen semantic classifier. Each row describes a metadata field rather than a
business record: property name, display label, entity name, neighboring property names, type,
annotation summary, locale, and the reviewed MockGen semantic hint.

The dataset intentionally excludes raw SAP Hub EDMX files, private SAP metadata, service keys,
paid-model transcripts, generated mock rows, and example business values.

## Files

- `train.jsonl`: project-authored base classifier training rows.
- `validation.jsonl`: project-authored base validation rows.
- `public-real.jsonl`: reviewed field metadata harvested from public API descriptions.
- `reviewed-fixtures.jsonl`: reviewed public OData and project-authored recovery fixtures.
- `provenance.json`: export manifest with source, split, license, and provenance metadata.

## Row Schema

Required fields:

- `property_name`: schema property identifier.
- `label`: human-readable label when available.
- `entity_type_name`: containing entity or schema object.
- `neighbor_properties`: nearby property names used as context.
- `type`: source schema type.
- `sap_annotations_summary`: compact annotation summary when available.
- `locale`: row locale.
- `hint`: reviewed MockGen semantic label.
- `source_service`: source service or synthetic fixture family.
- `source_kind`: provenance category.
- `license`: source/license note for the exported row.
- `provenance`: human-readable provenance note.
- `publication`: must be `public`.

## Intended Use

Use the dataset to reproduce MockGen classifier training, inspect label coverage, build
semantic-field classifiers for metadata-only mock generation, and evaluate out-of-domain
schema-property labeling. The rows are not end-user data and should not be used to infer
business records, customer records, or proprietary SAP service contents.

## Provenance And Licensing

Rows are exported by `scripts/release/classifier-dataset-export.ts` from an explicit
allowlist in `scripts/release/classifier-dataset-publication.json`. The exporter fails closed
when a row contains raw value fields, missing source metadata, private publication status, or
internal/proprietary source markers.

Some rows are project-authored under Apache-2.0. Rows derived from public API metadata are
metadata-only examples with source provenance preserved in `source_service` and
`provenance`; downstream users must respect the original source terms.

## Relationship To The Model

The paired model repository is:

https://huggingface.co/Unseen1980/mockgen-classifier-v1/tree/v1

The dataset repository is:

https://huggingface.co/datasets/Unseen1980/mockgen-classifier-data-v1/tree/v1

The current shipped classifier remains the accepted v1 artifact. Experimental retraining
runs are evaluated offline and only replace the published classifier when they improve both
service-holdout and domain-holdout benchmarks recorded in the model card.