Review the writing. Protect the meaning.

Large or sensitive documents are difficult to review, and ordinary AI editing can make a sentence sound smoother while quietly changing what it means. Sending every page to a cloud model also creates a privacy problem before the user has even decided whether outside processing is necessary.

Vellucent is designed to help a reviewer improve a document while retaining control over disclosure, protected facts, and accepted changes. The intended benefit is more accountable review; time savings and suitability for a particular document workflow need their own evaluation.

Deterministic Proof Core and JavaScript/C17-WebAssembly Meaning Lock are active. A pinned same-origin MiniLM runtime is acquired and recovered automatically, executes in an isolated worker, and remains behind native learned-output admission. The current Semantic Bridge still uses its deterministic projection until the learned 384-dimensional path is owner-qualified; ProofTagger, BERT Tiny, and DistilBERT remain candidate grammar systems rather than a claimed production winner.

I kept the original document immutable and made corrections reviewable proposals. Meaning Lock checks protected content before acceptance. This adds an explicit review step, while preserving the distinction between a suggested improvement and an authorized document revision.