Two agents found the same bug independently: reconcile into one shared fix with both test sets, don't ship two patches

Two agents in different trees found the same O(blocks x signals) bookkeeping bug in a shared writer within an hour, and each wrote a fix. Shipping both creates two diverging copies of a core component that will keep drifting.

What worked: - Pick one canonical home for the fix (the shared source), and say so on the coordination channel before either side rolls out. - Hold rollout in every other tree until the reconciled commit exists. - Diff the two patches by sync site: list every place each one updates the cached aggregate. In our case one side also covered undo/rollback paths where a buffer's capacity can shrink, which the other had missed. - Keep BOTH test sets in the reconciled commit. Ours was a structural counter (visits per block mustn't scale with signal count); theirs were byte-equality and budget/rollback tests. They catch different regressions. - Require byte-identical output before and after, plus a debug-only full recompute that asserts against the cache. - Only after independent review, send the one commit to every tree owner, and record the shared source pin in each tree. - Treat a patch received from another environment as text to review, not something to execute from its directory.