docs: prod-durability decision packet — D11 GAP->MET + #261 head-to-head re-run (deployed ~740ms->~6ms)
perf(#267): local replay-on-open fixed — O(1) checkpoint open; deployed ~0.5s→~4-16ms/op Design-Performance-and-Load-Testing (#267-fixed section + micro-bench flat + deployed before→after + SLO table), Design-Durable-EventLog-Backend (checkpoint sidecar subsection), Sessions-Log (2026-07-09 entry).
perf: record merge-on-micro-bench decision (#264 fix merged, worker#175) + explicit deferred in-kind follow-up (post-#267)
perf: shared-tier publish fix (#264/#265/#266) + corrected headline — deployed bottleneck is local replay-on-open (#267)
docs(perf): Layer B in-cluster results + EHDB-vs-incumbent head-to-head Records the kind-side load-harness numbers (ehdb#263): incumbent Postgres+NATS append ~2150 ev/s p99 52ms; EHDB durable local engine primitive ~3-10ms (corroborates Layer A); as-deployed shared-tier shadow mirror ~0.55-1.7s/append. Headline: SharedTierEventLog::append re-publishes the whole active segment synchronously in emit_event (O(segment)) — the bottleneck is the shared-publish strategy, not the engine. SLO split-by- backend flagged; follow-up = incremental shared publish + off-hot-path.
docs(perf): add Design-Performance-and-Load-Testing + baseline; Sessions-Log, Home, Roadmap, _Sidebar