Only reviewed factual labels, lexical roles, morphology and implementation-agent interpretations are exported. Original documents/corpus comments/definitions/images/code remain research sources outside the repository; no emotion annotation is lexical human-review certification. Raw HTTP, rendered entry observations and field serializations are explicitly distinguished, with previous failures and unknown versions retained. Existing source licenses and attribution remain. GoEmotions provider Google Research, repository Apache2 license and documentation are retained alongside public research sources; corpus text is not redistributed in dictionary outputs. Public-source spelling-only discovery searches were used; readings, costs, row IDs and dictionary payloads were not sent. No billing, keys, authentication or external account action.
