CHG-0016: Idle, the Screen on the Console, and No Buttons for a Record


Date	2026-10-12
Change type	Upgrade
Status	In progress
Systems	`nowplayd`, `lightd` and the console on the workload; the cover screen’s firmware; the tenant’s Terraform (one grant on the console’s account)
How	`make plan apply` in `infra/deevnet-tenant-eds`; `make -C deploy deploy WHERE=builder`; `make flash` in `firmware/cover-screen`
Risk	Medium. What lights the room changes: the stand dims by itself for the first time. Most likely to go wrong: the stand dim when something is playing, or lit when nothing is
Decisions	ADR-0024 and ADR-0025, which this is the evidence for
Related	CHG-0015, on whose branch this is built and whose console change this adds to

Summary

Three things, asked for together.

A track paused for fifteen seconds is no longer said to be playing. The console says nothing is, and the stand goes dim and warm.
The cover screen is shown on the console, beside the stand and the beacon.
While a record is playing the cover screen shows the sleeve and no buttons, whatever is touched.

Goal

iTunes paused: fifteen seconds later the stand is dim, and the console shows nothing playing and the stand dim.
Play pressed on the cover screen: iTunes plays and the stand is lit again.
The console lists cover-screen-01.
With the turntable current, a touch on the screen raises nothing.

Scope

In scope: nowplayd, lightd, the console, the cover screen’s firmware, the console’s broker account.

Out of scope: the stand’s firmware (the idle scene is an ordinary scene), the agent Pi and its drivers, the beacon, palette.

Risk and impact

Risk	Where	Guard
The console is refused its new subscription and does not start	the broker	the tenant is applied before the console is deployed
The plan changes something other than one grant	the tenant	read before it is applied: one account changed in place, nothing added, replaced or destroyed
The stand goes dim while something plays	the room	idle is decided only for a paused winner or none; a playing or held source overtakes it; `NP_IDLE_AFTER=0s` turns it off
The stand stays lit when lightd missed the topic being emptied	the workload	lightd takes five seconds of no cover after subscribing as none
The screen’s panel or touch settings are lost in the build	the board	the build’s settings were compared with the ones the board was last flashed from: both are the defaults
The wrong board is flashed	the Mac, with the stand and the screen both attached	the board on the port is asked its project name before it is written

Procedure

The tenant. make plan, read it, make apply.
The workload. make -C deploy deploy WHERE=builder.
The screen. make flash PORT=/dev/cu.usbserial-2120. Its credentials are not touched.

The screen and the workload are safe in either order: the old firmware ignores the new field, and the new firmware offers its buttons until told otherwise.

Verification

nowplayd’s log says “idle” fifteen seconds after a pause, and lightd’s a scene with idle=true.
eds/nowplaying/current/art is empty then, and the stand’s state says solid at 0.15.
The console: “Nothing is playing”, the stand drawn dim, and a pill for the screen.
The screen’s serial log says “no controls” for the turntable’s track.

Undo

Idle: NP_IDLE_AFTER=0s in deploy/nowplayd.env, and make -C deploy units.
The buttons: remove NP_NO_CONTROLS_SOURCES the same way; the firmware need not be changed back.
Everything: deploy the commit before this one, and flash the screen from it.

Outcome

When (UTC)	What happened
12:12	The tenant was planned: 0 to add, 1 to change in place (the console’s account, one subscription), 0 to destroy. The apply was left to its owner.
12:12	The board on `/dev/cu.usbserial-2120` said it was `cover-screen`, and was flashed with `6dab15e`. It started, found its credentials, connected, and read the current track: stopped, no cover. The stand, on the other port, was not touched.
12:18	Its owner had applied the tenant, and the services were deployed from the Builder (`049e1e6`). All four came back active. The console listed `cover-screen-01`, online, beside the beacon. `nowplayd` said it was telling controllers to offer nothing for the turntable.
12:18	What was current was the turntable, held and silent, its beacon on. So idle had nothing to act on: the track paused in iTunes was not the winner, a held source does not go idle, and the stand was lit at full from a cover still retained from before. Nothing was playing and the room did not say so.
12:23	A wait of its own for a source stopped and still current, five minutes, was deployed (`bed378c`). The turntable had been silent for hours, so it was idle at once: `nowplayd` said so, the cover topics were emptied, `lightd` sent the idle scene, and the stand reported taking it, solid at 0.15. The console showed no cover and the stand dim.

Departures

A second deploy, for the wait that was missing. The first left the room lit with nothing playing.
The screen was flashed before the tenant was applied and the workload deployed, not after. Either order is safe, and the board was at hand.

Follow-ups

The evidence the two decisions ask for.
A turntable left on with no record kept the last sleeve lit for as long as it was on, and was seen doing it within a minute of the deploy. Its owner chose five minutes, and it is NP_IDLE_AFTER_STOPPED.