sequence

The materialized sequence of all RgaIds in RGA order, including tombstones. Computed lazily and cached.

Exposed for WindowPolicy (in :kuilt-quilter) implementations that need to inspect the full ordered sequence (e.g. WindowPolicy.byCount).

A mutation that can prove the result without recomputing threads it forward through RgaCache.sequence (#2193), so a chain of appends pays one computeSequence() rather than one per turn. Every other path leaves it null and this recomputes once.

Only a threaded order propagates, not merely a materialized one. The chain is armed by removeAt / removeFirst / insertAt, which force this lazy and thread the result on; an instance whose sequence was forced by a plain read carries nothing in RgaCache.sequence, so insertAfter on it threads nothing and the next read recomputes. An append-then-read loop therefore still pays one computeSequence() per turn. That is unclaimed value rather than a correctness problem — the guard's premise holds equally for a forced lazy — and closing it is tracked by #2225.