Every rule in the mapping registry, read only. Editing happens where an editor exists.
Three different facts, and they are worth keeping apart.
Stored means the row is in the registry. Active means its status is active and it carries a published revision.
Read by an engine means some code actually loads it and acts on it. A rule can be stored and active and still be read by nothing,
which is the case for most of what is here.
Four MWE rules are missing from this registry, and that is a gap, not a design.
Eligibility, discovery, ranking and correction policy are every bit as much rules as the one below: they decide which
lexicon entries qualify, which word sequences become candidates, how those are ordered, and how a human correction applies.
Today they are hardcoded, in candidates() and the SCORING policy in
assets/citations-render.js and in the pack build. Hardcoded means nobody can inspect the policy, change it,
preview the change, or reverse it.
The requirement is that every automated MWE decision has an inspectable policy, human controls, an explanation of its
effect, and a way to correct the result, with changes previewable, versioned, tested and reversible. The engine stays code.
The policy does not get to be untouchable because code implements it.
One consequence you can see today: editing the term lookup prompt below cannot affect which phrases are discovered,
because discovery runs first and has no editable policy at all. Making discovery and ranking editable is required work.
See DR-094.