the canary isolation matrix resolved the static-binary investigation: - "musl breaks RocksDB" is retired — musl builds/links/runs SIGILL-free on 0.16 at thread scale (was a 0.15 codegen artifact). - mimalloc is ruled out for zlay's ~2,800-thread model — RSS climbs unbounded (v2/v3/purge-forced/even under glibc); the glibc+mimalloc row isolated the allocator as the variable. so a static-musl deploy is blocked on the allocator (glibc malloc's page-return is what keeps prod flat and is glibc-specific), and there's no need anyway. - musl-investigation.md: full matrix + verdict + dead-end allocator list + where the revivable -Duse_mimalloc harness lives (musl branch, not merged). - CLAUDE.md / deployment.md: correct the stale "must use glibc / musl breaks RocksDB" notes to the evidence-backed verdict. - design.md: note mimalloc evaluated + ruled out in the memory model. - relocate the three 2026-06-12/14 operator handoffs into docs/handoffs/. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>