Use a 65536-page checkpoint threshold during incomplete full all-author replay. Reapply the connection-local policy from durable index state after preparation and process restarts. Restore the normal 1000-page threshold before live readiness. Keep normal startup, subsets, participant replay, and completed sources on the normal threshold. Preserve schema, cache configuration, atomic checkpoints, and synchronous=NORMAL. Larger intervals reduce repeated database page writes. Local benchmarks improve new-record throughput by 13% and update throughput by 25%, with the final checkpoint cost included. Document WAL growth, longer stalls, and the larger power-loss replay window. Race tests, vet, SQL checks, both Nix builds, and module checks pass. tsk-47