Review policy
Design readings have no external ground truth, so review status is a first-class category here, enforced in the content itself rather than only in a policy document — to keep invented content out of published entries. Method
Reviewed
Pending
Candidate
Epistemic labels
| Label | Meaning |
|---|---|
| Axiom | A premise of the system's position, an input to research and teaching, not an empirically established conclusion |
| Method | An established methodological norm the project follows |
| Measured | A quantitative result backed by computation or data, naming the side measured (machine / human) |
| Done | Completed work or a live asset |
| Planned | Work begun but not finished |
Admitting a new sense
A new sense of a component or stroke enters the database only if it meets all four conditions: it recurs independently across characters; its extension can be derived; it has a structural source; and it makes a prediction a third party can test.
Where AI stands
AI produces candidates, never the ground truth of a design reading. AI-suggested readings sit in the database's review queue as candidates, with consistency and ablation metrics, and only the author's review can make them reviewed. The AI study assistant answers strictly from reviewed entries and always distinguishes the three statuses.
Corrections
To report an error, tell us the page, the paragraph and your grounds through the contact page. Corrections are noted in the text with date and reason; a correction to a design reading (such as assigning “reaching upward” to 由 rather than 甲) also updates the database and the teaching units concerned.
