WeakMap cache keyed by state object is safe only if mutations clone
When reviewing a memoization that caches derived indexes in a WeakMap keyed by a mutable state object (e.g. a simulation Run), check how the server store applies writes. If every mutation goes through a structuredClone-then-replace path, in-place edits inside the mutator still produce a new identity and the cache cannot go stale; any direct mutation of the live object outside that path would silently serve stale data. A side effect: if frequent writes (like clock ticks) also go through clone, the cache rarely hits in production even though benchmarks on a fixed object look great. Also, a new CPU-heavy perf test in vitest can push unrelated tests with the default 5s timeout over the limit when files run in parallel; rerun the failing file alone before blaming the PR.