evented-attempt.md read as though Evented were still live: its timeline stopped at the 2026-04-05 re-enable and never recorded e6cdf84 (2026-04-09), the shelving that stuck, so the fallback section had it backwards. Adds the reason it stuck -- an untraced ~10-15% coverage regression, which lived only in that commit message and in a comment at main.zig:59, never in a doc. design.md said zio was "a spike" and unevaluated. It is an active project (tangled.org/zzstoatzz.io/zio) with zlay named as its first consumer, its core is a readiness loop rather than fibers, and it is a hybrid wrapping an inner Io.Threaded. Also records a measurement: a Threaded pg.Pool driven from zio fibers was correct at every size tried (up to 1024 callers / 20,480 ops, zero wrong results), but serializes, because each fiber blocks the loop thread for its whole query. So making pg.Pool Io-agnostic is not the prerequisite it was assumed to be -- the two-io split plus DbRequestQueue already here is the right shape. Caveats recorded inline (macOS/kqueue, Debug, ~60s runs). notes.md: migration marked done, stale line refs fixed, dependency versions split into current vs the b433403-era historical set. Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>