# Upgrade from main 9c3d5d6 (the v1.0.0 candidate) to 1.0.1 by binary swap (2026-10-08)

On the one-server node of `docs/evidence/installer/one-server-e2e/` (Ubuntu 24.04, systemd; Postgres 16, ClickHouse
26.3.39.7 + Keeper, PeerDB stable-v0.37.10). Raw output; home paths replaced with `~`; `SHA256SUMS`.
Both binaries print `chlift dev` (only release.yml stamps a version); `binaries.sha256` gives each one's source commit:
**R** = main 9c3d5d6 (the v1.0.0 candidate, not tagged yet), **Q** = the 1.0.1 queue.

- `R-install.log`: install re-run after a restart of the node: every check ok.
- `R-plan.log`, `rel.yaml`: `public.users` (22688 rows) planned `ORDER BY (id)`: the planner fix is on main.
- `R-start.log`, `R-status.log`: mirror chlift_rel RUNNING, postgres ~22688, e1 22688.
- `cdc-rel.sql`, `cdc-rel.log`, while R ran the mirror: INSERT 2000, UPDATE 1899 (rewrites updated_at), DELETE 207; pg_rows 24481.
- Binary swap to Q. `Q-install.log`: install re-run, "clickhouse-server running". `Q-status.log`: the mirror R made is
  RUNNING (raw e1 count 26561 before ClickHouse merges row versions).
- `Q-check.log` (`migrate check -checksums`, by Q): postgres 24481 rows; replicas e1:24481; 64 buckets x 5 columns, 0 differ.
  Its exit code was not recorded; the printed result is the evidence.
- `Q-abort.log`: mirror, peers, ClickHouse database and publication dropped; "mirror slots left: 0".
