What the tool does
The tool accepts a short, general phrase to help locate a listed topic. Matching runs in the browser and converts a reliable match into a fixed selection ID. When independently enabled, the server makes one provider call per API request. It supplies one to four scrubbed public units under request-local IDs u1 through u4 plus fixed ordering instructions. The provider receives no typed phrase or other user prose, source metadata, or server-owned display text. It returns only strict { "schemaVersion": 1, "orderedUnitIds": [...] } JSON containing an exact permutation. The server independently requires every ID exactly once and fails closed on any mismatch. It then renders immutable, prewritten review-stage bilingual unit text, each with one prebound official citation. The title, generic summary, caution, citations, and all citation metadata are server-owned. No model-authored legal prose is ever displayed.
What leaves the device
The request contains only the selection ID and locale. The typed phrase, names, dates, amounts, documents, and case narratives do not cross the API boundary. The tool does not place the question or result in the URL, browser storage, or analytics.
Limits and availability
The list is limited to topics with prepared context and source records. An unmatched input stops locally. The tool does not assess an individual situation, choose a procedure, predict an outcome, or draft a document. Its output is general legal information, not legal advice. This page remains in review, and the model connection stays disabled until the release, privacy, review, and evaluation gates are complete.