Top: one beat — graphs, numbers and the process table change on the same tick

Graphs stepped once a history point (1 s at a 5 min window) while the numbers changed every
tick; they now slide a fraction of a column each tick. The process table (and the GPU boxes,
MHz and watts it carries) was polled on its own timer; it is now read per sample and committed
with it. GPU clients are walked every sample, and the 1–2 s floors on the gallery, Network,
Docker and ZFS ARC live views are gone.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

BUGS.md

@@ -752,3 +752,27 @@ before its own login).
     404 — harmless, but a stray.
 12. **Guest logins can open Users and Settings** — by design ("a guest can do everything you
     can"), but worth a second look now that Users includes removing accounts.
+
+# 2026-10-23: the rate obeyed, but not in step
+
+**Graphs and numbers on Top moved on different beats, whatever the rate.** Every machine was
+sampling at exactly the set rate (measured: 500 ms between samples on all five nodes), so the
+sampler was not the fault. Four things in the console were:
+
+1. ✅ **Graphs stepped once a point, not once a tick.** A point is the history window split 300
+   ways — at 5 min that is 1 s — so at 250 or 500 ms the numbers changed two to four times for
+   every move of a graph, and a narrow box (phone, net graph) moved once a *column*, every few
+   seconds. The columns now slide by a fraction of one on every tick (`columnsAt` in
+   `graphcols.ts`), the newest growing in from the right; a finished column still never changes.
+   The per-core strips keep the last 18 ticks rather than stride points.
+2. ✅ **The process table, MHz, CPU watts and every GPU box ran on their own timer** — a
+   `setInterval` on `/api/procs` with an arbitrary phase against the sensor stream, so half the
+   page changed up to a tick after the other half. The table is now read on each sample and the
+   two are committed together: one render per tick for the whole page.
+3. ✅ **GPU clients were walked at most once a second** (`gpuScanEvery`), so per-process GPU% and
+   the engine meters lagged the device's numbers. Walked every sample now.
+4. ✅ **Hidden floors on other live views:** Sensors gallery (1 s), Network live rates (1 s) and
+   speed-test needle (0.4–1 s), Docker live (2 s), ZFS ARC (2 s). All at the rate now.
+
+Not changed: container status sensors are read every 15 s (`docker ps` plus an inspect) and
+filesystem usage every 30 s — slow-moving facts behind an exec or a statfs of every mount.