# Licenses and attribution

This is a collection with separately licensed components, not a newly trained model.

| Component | Attribution | License |
| --- | --- | --- |
| Simple English article snapshot | Wikipedia contributors; distributed by Wikimedia | CC BY-SA 3.0, as offered by the pinned source dataset |
| Chunks, catalog, extracted assertions and Wikipedia vector index | Derived from that snapshot by Commonplace | CC BY-SA 3.0 for the adapted Wikipedia material |
| EmbeddingGemma 2 files in `encoder/` | Google DeepMind, original upstream notices retained | Apache-2.0 |
| Empty memory schema in `seed/` | Commonplace contributors | Apache-2.0 |

The upstream Wikipedia dataset also offers GFDL; this distribution uses its
CC BY-SA option. See the preserved [upstream card](provenance/UPSTREAM_DATASET_CARD.md).
Full license texts are in [licenses/](licenses).

Source identity and article attribution are preserved in the Parquet data and
SQLite catalog. Each article has its original Wikipedia URL; its contributor
history is available from the article's History tab. For automated attribution,
retain the stored title, URL, source hash, and applicable license alongside
returned text. Do not attribute model interpretations to Wikipedia contributors.

The Simple English snapshot was cleaned by the upstream dataset, split into
overlapping source-offset passages by Commonplace, encoded into normalized
512-dimensional vectors, and indexed for lexical/entity lookup. Narrow extraction
rules derived relationships with source quotations. These are changes to the
original representation, not endorsements by Wikimedia or Google.

The encoder is the unmodified pinned upstream payload. The software loading it
disables vision/audio configurations for this text-only use; no weights were
fine-tuned. Its original model card is retained at [encoder/README.md](encoder/README.md).

There is no blanket software license over all data in this collection. The
Commonplace software repository's Apache license does not replace these terms.