CHG-0017: The Gap Between Tracks, and a Silent Turntable


Date	2026-10-12
Change type	Upgrade
Status	In progress
Systems	`npagent` on the agent Pi (`np-relay`); `nowplayd` and the console on the workload
How	`make -C deploy/relay push`; `make -C deploy deploy WHERE=builder`
Risk	Low. Most likely to go wrong: a wrong title for a few seconds after a quiet passage, or a room that dims while a record is being turned over and surprises someone
Decisions	ADR-0026, which this is the evidence for
Related	CHG-0013, CHG-0016

Summary

Three things about silence, asked for together.

The turntable driver listens for the gap between two tracks, and names the track after it from one clip. A track change took thirty to sixty seconds to show.
A turntable that has gone silent is idle about twenty seconds later, not five minutes: the stand dims and the sleeve is taken away.
The console stops showing a sleeve under “Nothing is playing”. That one is a mistake put right, not a decision: the page hid the picture in a way its own style sheet overruled.

A track that follows a gap is named about thirteen seconds in.
The needle lifted: about twenty seconds later the stand is dim, and the console and the cover screen show no sleeve.
The needle down again: the sleeve and the lights are back with the first track named.

In scope: the agent on the Pi; nowplayd’s one setting, NP_IDLE_AFTER_STOPPED; the console’s page.

Out of scope: lightd, palette, every firmware, the tenant. The idle scene is the one the stand already knows.

Risk	Where	Guard
The gaps on this input are not as quiet as -55 dBFS	the driver	then no gap is heard and the driver behaves as it did; nothing is worse
A wrong name after a quiet passage	the console	it is one answer and the next clip corrects it; the cover is held by the album and the collection
The agent does not come back after the push	the Pi	`make -C deploy/relay status` straight after; the push before this one is the undo
The room dims while a record is turned over	the room	that is what was asked for, and it is a setting

Either order: the two halves do not depend on each other.

The agent’s log names a track about thirteen seconds after a gap.
With the needle up: nowplayd says “idle” about twenty seconds after the last sound, eds/nowplaying/current/art is empty, and the stand’s state says solid at 0.15.
The console says “Nothing is playing” with no sleeve above it.

The gap: push the agent from the commit before this one.
The lights: NP_IDLE_AFTER_STOPPED=5m in deploy/nowplayd.env, and make -C deploy units.
The console: deploy the commit before this one.

When (UTC)	What happened
21:36	The agent was pushed to the Pi by its owner (`65e6dfc`); the push was refused to the assistant. It came back active with both drivers online, listening to `plughw:UA4FX`. The turntable was silent, so nothing about the gap was seen.
21:38	The workload’s deploy, run by its owner from the worktree, built and loaded the four images and then stopped: the tenant’s directory was not passed, and the worktree has none. Nothing was installed or restarted.
21:39	`make units`, with the tenant’s directory named, installed the settings and restarted the four services (`2b486a2`). All four came back active. What was current was iTunes, paused long before, so `nowplayd` was idle at once, `lightd` heard no cover and sent the idle scene, solid at 0.15. `nowplayd` said its wait for a held, stopped source was 1s. The turntable was not the current source, so that wait had nothing to act on.

Both halves were run by the owner; the assistant was refused the push.
The workload was deployed from a worktree, which has neither the tenant’s backend settings nor discogs.env. The first was named with INFRA=, the second copied in. The first try lost INFRA= to a line break.