Review Architecture

Kinesema uses review state to prevent provisional output from appearing more authoritative than its evidence supports. Parser research continues privately; proposed language output still requires human review and is not automatically ready for motion. The website no longer exposes an interactive parser.

That status does not mean a reviewer approved the result. It means the result must not be treated as accepted translation or motion-ready input.

Review State and Review Action

Review state is information recorded by the system. A review action is a judgment made by an identified person or a bounded technical check.

These must remain separate. Setting a field to reviewed cannot manufacture a human decision, and passing a schema test cannot establish linguistic correctness. A legitimate transition requires the evidence and authority defined for that exact subject.

Different Questions Need Different Authority

Kinesema may need several kinds of review, depending on the material.

Technical Validation

Technical checks can verify response shape, deterministic behavior, field allowlists, privacy boundaries, hashes, test expectations, and refusal paths. They can show that software followed a contract; they cannot approve SgSL.

Linguistic Review

Qualified human review may assess meaning, grammar, sign choice, naturalness, context, or non-manual requirements. The reviewer’s language competence and the exact material reviewed must be recorded. One bounded judgment does not become universal language authority.

Motion Review

Future motion work may require checks of pose, direction, timing, signing space, transitions, and physical behavior. A technically valid animation may still be linguistically wrong, while a linguistically supported plan may still contain a rendering defect.

Accessibility Review

Accessibility review considers whether people can perceive, navigate, understand, and operate the interface through relevant access paths. Clear static text is useful evidence, but it does not by itself establish keyboard, screen-reader, reflow, contrast, localization, or complete product accessibility.

Community and Cultural Guidance

Deaf community members may provide lived, cultural, naturalness, teaching, or usage perspectives that technical tests cannot supply. Their contribution must be voluntary and scoped, and no participant should be presented as speaking for an entire community.

Product and Publication Decisions

Even reviewed material does not become public automatically. Product inclusion, publication, hosting, data collection, and runtime activation remain separate owner decisions with their own safety and privacy checks.

Minimum Decision Record

A useful review decision should identify:

• the exact subject and version reviewed;
• the question the reviewer was asked;
• the evidence or presentation they received;
• the reviewer’s relevant role or competence;
• any role overlap, limitation, or material AI assistance;
• the outcome and required correction;
• what the decision does not establish; and
• the date and resulting state.

The record should preserve the reviewer’s actual judgment rather than replacing it with a stronger project-authored conclusion.

Possible State Transitions

Future review records may need more detail than a single Boolean field. A subject could remain proposed, require correction, be blocked, be accepted for one bounded use, or be superseded by a later version.

These are design directions, not a claim that every state is implemented in the current parser. Any future state model should define which evidence and authority permit each transition and should fail closed when those conditions are absent.

Refusal Is a Valid Outcome

A review system should be able to stop a result rather than forcing every item toward approval. Refusal may be appropriate when evidence is missing, the reviewer lacks the required authority, the subject changed after review, or the requested use exceeds the accepted scope.

Visible refusal is safer than silently filling a gap with generated language, placeholder motion, or an unrelated review decision.

Relationship to Community Review

The software architecture can preserve subjects, versions, decisions, and boundaries. It cannot create community participation or linguistic authority.

The Proposed Community Review Process describes how future human participation could be organised. The Parser Research Boundary documents the current review-required output, while the Motion Research Direction explains why motion requires additional evidence and review.

What Review Establishes

A completed review establishes only the recorded outcome for its stated subject, scope, evidence, reviewer role, and date. It does not prove universal truth, permanent correctness, community consensus, product accessibility, or safety for every use.

Kinesema’s goal is not to attach more approval labels. It is to make the real authority behind each decision visible and to keep unsupported transitions blocked.