AB updated Timers from the catalog and LTC died: the update replaces the
module folder wholesale, which deleted the compiled ltc-capture binary —
and AB's Xcode CLT cannot rebuild it (the SwiftBridging modulemap defect;
the module's new self-build correctly runs there and fails with exactly
that error). The binary on AB has always been compiled elsewhere and
shipped, so 'update' was guaranteed to break LTC on that machine.
installModule now carries executables (any execute bit) the download does
not include over from the copy being replaced — or from the app's bundled
copy on a first install — preserving path and mode. Executables only, on
purpose: those are the gitignored per-machine build artifacts (ltc-capture,
spl-capture), while stray text files from an old version — code an update
deliberately removed — must not be resurrected. The Timers module's
staleness check still notices a carried binary older than its updated
source and rebuilds behind it where the toolchain works.
Verified end-to-end against the real Gitea catalog: an installed 1.4.9
copy holding a fake executable and a stray .js file, updated to 1.5.0 —
the executable survives with its mode and content, the stray file is gone,
and the log names what was kept.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
AB updated the Timers module to 1.5.0 and its new LTC capture-tool
self-build never ran: mountModule re-required the module's server entry,
and Node's require cache handed back the OLD code — the catalog update
had swapped the folder on disk, but the process kept executing what it
loaded at startup until ProdDash itself restarted. The manifest, read
fresh from disk, meanwhile reported the new version, so the admin page
claimed an update that functionally hadn't happened. Every module update
to date has silently depended on a follow-up app restart.
mountModule now drops the module directory's entries from require.cache
before requiring, so every (re)mount runs the files on disk — nested
requires (ltc-listener.js and friends) included. Modules already must
come up clean on re-init per the module guide, so config-change remounts
see the same contract, minus the trap.
Verified with a probe module in a scratch data dir: change a nested
require on disk, PUT config to remount, the route serves the new code
(it served the stale code before the fix). The Timers module re-inits
cleanly across repeated fresh-code remounts.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
A catalog install ships source only, and a catalog update deletes the
module folder — a previously built ltc-capture binary with it — so every
install or update left LTC dead until someone compiled the tool by hand.
The build also still used the blocking spawnSync that once froze the whole
server (the bug spl-meter fixed in 0.4.0), and only ran when a listener
started: no tool meant no device list, and no device meant no listener.
ensureCaptureBinary() (module scope in ltc-listener.js, the spl-meter
0.4.2 pattern) now builds the tool in the background as part of coming
up: kicked when the LTC source starts and by the admin device picker,
single-flight, built to a temp name and renamed so a killed build can't
leave a truncated binary, and rebuilt behind a still-serving binary when
a module update leaves the source newer. The waiting listener starts
itself the moment the build lands.
Tile-side warnings are now one line — 'Not configured — see Admin',
'Unreachable', 'Capture tool building…' — with the client's source-label
prefix no longer repeated. The full diagnostics (host:port, errno, the
capture tool's own words, build instructions) move to the admin health
line and the log, where the person who can act on them is looking.
Verified: build from the exact catalog-install state (source only) lands
and lists 12 real devices; touching the source triggers an atomic rebuild
while the old binary keeps serving; /ltc/devices kicks the build and
answers with the device list; live server shows terse source messages on
the tile and the detailed story in /api/admin/state health. ltc and phase
selftests still pass.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Three things, the first on its own.
**Auto-scroll moves the tile, not the board.** `scrollIntoView` walks up and
scrolls every ancestor that can scroll, so bringing the current item into
view dragged the whole dashboard with it whenever the board was taller than
the window. The list's own `scrollTop` is moved instead, measured from
rectangles because the rows' offset parent is the tile rather than the list.
**The current item fills as its song plays.** When an AbleSet module is
installed, configured and playing a song that IS the item on screen, that
item's background fills left to right with the song. The song is tied to a
plan item by the same name matching the live ProPresenter document already
uses, so a bar can never quietly fill the sermon while the band rehearses
something else.
**Its sections slide past, bottom right of that item.** Every section of the
song as a strip, the played ones dimmed, the one playing filling in turn and
kept centred — by moving that strip's own `scrollLeft`, for the same reason
as above. The chips are rebuilt only when the song changes; moving to the
next section just moves the classes, so the strip keeps its place.
Both are switches in the tile's gear, **Song progress** and **Song
sections**, and both draw nothing at all unless there is an AbleSet module
to follow: the server reports `available: false` for no module, a disabled
one, or one with no host, and re-checks every 15 s so installing AbleSet
later lights it up without a restart.
The server follows the AbleSet module over loopback exactly as it follows
ProPresenter's, and republishes a small slice — song, sections, playhead in
beats with a stamp — as its own `ableset` SSE event. Its own event because
it moves five times a second and must never drag the plan (items, teams)
onto the wire with it; the tile carries the playhead forward from the tempo
between packets.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
ProPresenter reports running timers as truncated whole seconds, and the
poller re-anchored the tiles' local count at the bottom of that second on
every 500 ms poll — so the displayed digit ran up to a second ahead and
got yanked back a step whenever a fresh poll sampled the next flip later
than the last one did. Tiles flickered ...12:33 → 12:34 → 12:33... about
once a second.
phase.js turns the quantized reports into one stable anchor per timer:
the only information in a poll is a FLIP (the report stepping one second
in the direction of travel), which dates the true boundary to the
midpoint of the two polls that straddle it. While reports stay consistent
with the anchor (±450 ms) the anchor holds, so tiles count along one
steady clock; a report the anchor can't explain (reset, seek, resume,
accumulated drift) re-anchors. Timers carry the anchor as a per-timer
asOf the client extrapolates from (optional in the provider contract,
documented in the guide); the client additionally refuses to tick a
running card backwards over sub-1.5 s jitter.
phase.selftest.js (66 checks) drives a simulated ProPresenter — truncated
reports, jittered polls — through the tracker and the client's render
math: digits never step backwards, dwells hold 1 s (the ~2 s zero-crossing
dwell is trunc's honest behavior), anchors stay within tolerance of the
true value, resets snap, pause/resume re-locks, 0.2% clock drift is
absorbed. Verified live against tools/pp-timers-mock.js: 78 s of 100 ms
DOM sampling across four running cards — every step ±1 in the card's own
direction, dwells 998–1002 ms, zero regressions; upstream kill/restart
still recovers.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Colors and Status colors were two groups asking the same question, so they
are one: team header, the tints and the three status colours together.
The Positions switches go entirely — the team switches already decide what
a tile shows, and "Group by position" still arranges what is left. With
them go GET /positions and positionOptions().
Two rules the tile was breaking on the way past: a gear group may carry no
`help` (rule 5), and only hide-toggles belong in Show (rule 8) — so the
group blurbs are gone and the Columns picker sits loose at the top beside
the team switches.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The module now offers two tiles in the ＋ picker, Plan and Teams, sharing
the one Planning Center connection and the one poll.
Teams draws the current plan's team and position assignments: each team as
a band of its people, with the person's photo, their position and a dot for
confirmed / unconfirmed / declined. Photos are proxied and cached by the
module (GET /team-avatar/<planPersonId>) so no browser ever talks to
Planning Center; a person with no photo gets their initials.
Everything shown is the tile's own choice, so two tiles can differ:
- Teams and Positions — a switch per team and per position, all on until
someone flips one (GET /teams, GET /positions feed the lists)
- Colours — a picker for the team headers, or Tint teams to give each its
own, optionally carried into the rows, plus the header fade and pickers
for the three statuses
- Show — twelve switches (top bar, team headers and counts, avatars,
positions, group by position, status, notes, declined and unconfirmed
people, teams nobody is on, columns)
Server: team_members?include=team,person rides a slow lane (re-read on a
plan change, else at most once a minute) and can never fail the plan read —
a lost roster keeps the last good one and says so. The service type's own
team list is read too, cached ten minutes: Planning Center only returns a
team once somebody is scheduled on it, so that list is the only way to show
a team nobody has signed up for.
No shell changes: picker entries carrying their own instanceSchema, the
colour field and multiselects all already exist.
tools/pco-mock.js grows team_members, the service type's teams and stand-in
avatar images, including a declined person, people with no photo and a team
nobody is on.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Yesterday's version of this was a Python script under `modules/` with a
LaunchAgent to keep it alive — which the guide rules out in as many words:
everything a module does runs inside its server entry, no companion programs
or daemons an operator starts by hand. It also could not put anything on a
dashboard, which is half of what it was wanted for. Both halves are now in
the right place.
`modules/resi-cues/` is an ordinary module. Its tile is one number as large as
the tile allows — the countdown to the next Resi cue, the cue's name above it,
the playhead and the cue just passed underneath, and the whole cue list behind
a tile switch. It also declares `"provides": ["timers"]` and serves `/timers`
and `/timers/stream`, so the countdown and the playhead turn up among the
ProPresenter timers in the Timers module without either side knowing about
the other.
The countdown ticks in the browser. The server sends the playhead with the
instant it was true and whether playback is running, and the tile adds the
elapsed time since; the Timers module extrapolates from `startedAt` the same
way. So a snapshot only has to be sent when something structural changes — a
new cue, a pause, a seek — rather than at 1 Hz to every open dashboard.
`resi.js` holds the part worth reading twice. Resi republishes a decoder's
position only every few seconds, on an irregular cadence, and each reading is
already stale when it appears; a countdown built straight from it sits seconds
behind and lurches. So every reading is kept as media time minus wall time and
the largest recent value is trusted — the reading that arrived with the least
delay — which a slow round trip cannot drag backwards. A reading outside the
link's measured noise floor counts as a seek: drop the lock, snap, and re-read
the cue list at once instead of on its timer. Playback is only called paused
once the playhead has been still for longer than Resi's own cadence, because a
fixed timeout mistakes a slow update for a stop and freezes the countdown
mid-service.
Driving the ProPresenter timer on the stage screen is Companion's job now. It
is already a service on that machine, so nothing new has to be kept alive, and
this module no longer has to reach ProPresenter at all — it only has to be
right about Resi, which is the hard part.
`resi.test.js` drives the estimator on a fake clock against nine simulated
Resi profiles — cadences of one to twelve seconds, staleness to three,
irregular timing, variable round trips — covering lock-on, a fired cue, a seek
backwards, a pause, and a pause longer than the offset window. A punctual
fixture proves nothing here: against one, three separate breakages of the
estimator go unnoticed. `server.test.js` runs the real server entry against a
canned Resi account and checks what the two consumers actually read, including
the `/timers` shape field by field. Every mutation tried against either turns
them red.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The ask was a Stream Deck button showing the time until the next Resi cue.
What it turned into is simpler: ProPresenter already renders countdowns, and
it already has an HTTP API, so nothing has to sit in between except something
that knows when the next cue is.
`resi_to_propresenter.py` watches a decoder's playhead and keeps a
ProPresenter countdown timer armed with the seconds until its next cue, with
the cue's name as a stage message beside it. Both halves of that come from
Resi's cloud rather than the decoder — the decoder reports its position
upstream and `central.resi.io` hands it back — so there is nothing to install
at the decoder end and nothing polling it every second.
The playhead is estimated rather than repeated. Resi republishes `position`
only every few seconds, irregularly, and each reading is already stale when
it appears; sent straight through, the countdown sat three seconds behind and
lurched. So each reading is kept as media time minus wall time and the
largest recent value is trusted — the reading that arrived with the least
delay — which a slow round trip can no longer drag backwards. A reading
outside the link's measured noise floor counts as a seek: drop the lock, snap,
and re-read the cue list immediately instead of on its timer. Playback is
only called paused once the playhead has been still for longer than Resi's own
cadence, because a fixed timeout mistakes a slow update for a stop and freezes
the countdown mid-service.
This is not a tile. There is no `module.json`, so `readManifests` skips the
folder and it never reaches the ＋ picker; it lives under `modules/` because
that is where the production tooling lives. That is also why it carries a
README when no real module does — there is no *Included modules* entry it
could have instead — and the repo README says so where the module list ends.
`install-agent.command` sets up a LaunchAgent so it runs from login and
restarts if it stops. It deliberately runs a copy in Application Support: a
login agent cannot read `~/Documents` without Full Disk Access, and macOS
answers "Operation not permitted" rather than asking. Settings live there too,
outside the checkout, the same way module config does. The Resi password is
read from the Keychain and never written to disk; with none set the agent
says so once and waits rather than relaunching in a loop.
`test_playhead.py` drives the estimator on a fake clock against nine
simulated Resi profiles — update cadences of one to twelve seconds, staleness
to three, irregular timing, variable round trips — and six of seven deliberate
breakages of the estimator turn it red. `test_end_to_end.py` runs the real
loop against a real ProPresenter with a simulated decoder, and has: it armed
for the first cue, counted down in step, rolled over to the second and stopped
when the cues ran out. Neither covers the live Resi calls, which need real
credentials.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Two things were out of place. `modules/band-lineup/README.md` was the only
README inside any of the eight modules — the guide's folder shape is manifest,
client, server, assets, and every other module is described in this README's
*Included modules* instead. It was also describing a module that no longer
exists: photos and birthdays "come from Planning Center", guests "pinned per
slot", and a refresh interval the manifest has never had.
The entry here had drifted the same way, calling the Drive folder somewhere
to put "manual photos" and the board something Planning Center fills. It is
the other way round now, and has been since 1.10: the board is shared server
state and the source of truth, the photo library fills it, and Sync is
auto-fill that only touches slots nobody placed by hand. The entry says that,
and picks up the Choir spot and the title-bar columns while it is there.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Four campuses run different Sunday schedules and take their resi feed from one
of two broadcast origins. When a service goes wrong — a machine crashes, a feed
is unusable — the booth has to know which service at which campus can be
altered, and which campuses are riding on the one that broke. That lived in
people's heads and in messages.
The times come from the church website rather than from configuration, because
the site is what stays current when services change through the year. Each
campus page carries a "Times" block; the module reads it and re-reads it on a
timer. Two things on those pages are not services and are excluded on purpose:
the main-office hours in every page's footer, and the "Office Hours" block
Apollo Beach carries inside its own Times section. The parse is bounded to the
Times heading and stops at either.
A mark on a service — issue, replay, do not use, with a note, and for a resi
campus which of the origin's services it is taking — is the booth's shared
answer, so it lives in the module's server state and reaches every tile over
SSE. Per-tile settings would let two campuses disagree about whether the 9:30 is
going out. Marks are written to the data directory so a restart mid-morning
does not lose them, and they expire after 36 hours so last week's never greet a
new Sunday.
Campuses added by hand in admin are a stopgap for a site that has not caught up
yet: as soon as the website publishes that campus, the site's times take over
and the admin health line says the manual line can be removed.
Verified: the parser against all five live campus pages, and 31 unit tests plus
16 client checks, each proven to fail under mutation — both parse guards, the
merge rule, dependent recording, and three client behaviours. The suite is run
here under JavaScriptCore, since this machine has no Node.
Not verified here: the module guide's resilience checklist needs a running
ProdDash — upstream loss and recovery, two tiles side by side, and tile removal
leaving no timers behind.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The reference fork had drifted four builds behind the app actually running
backstage. What it was missing is Birthday Mode, the feature the ProdDash
module was ported with and this snapshot never had:
- main.js grows the Planning Center birthday fetch (`pcoFetchBirthday`),
mm-dd normalising, and a `pco-test-birthdays` IPC handler behind the
settings pane's Test Birthdays button.
- renderer.js gains the whole celebration: the Wednesday→Tuesday birthday
week, the canvas confetti loop, the gold glow, and the controls that switch
it on and set its intensity.
- index.html and styles.css carry the Birthday Mode block in settings.
Purely additive — 597 lines in, four out, and those four are the version and
build number. Source only, as the fork has always been: `.claude/` settings
stay as they are (the local copy's are one machine's accumulated permissions,
and it has no launch.json to give), and the stray test file beside them is not
part of the app.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Every gear change remounts the tile, and the remounted tile joins the
page's shared stream mid-flow, so it heard nothing until AbleSet next
changed — with the transport stopped, never. Tiles went blank until a
reload. The tile now fetches /state when it starts and whenever the
stream (re)opens.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A new module that subscribes to AbleSet's OSC API (UDP 39051) and streams
what it is playing to ten tiles: Now Playing (song, tempo, meter, duration,
time left, section, progress bar, next song, songs left), Song Sections
(a sliding strip, or a lottery wheel from the title bar), Setlist,
Transport, Tempo, Song Time, Set Time, Section, Next Song and Status, each
coloured by its state. The song and set countdowns are offered to the
Timers module.
A heartbeat watchdog resubscribes when AbleSet goes quiet (a restarted
AbleSet forgets its subscribers). Time left falls back to the song length
and playhead when AbleSet stops sending it (Live disconnected).
tools/ableset-mock.js plays a setlist over the same API for testing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The spectrogram scrolled by copying the visible canvas onto itself one step
along. That canvas already held the frequency labels and gridlines, because
they were drawn straight onto it, so every frame smeared them sideways and
left a trail. Arrived with the labels in 0.3.0; the gridlines did it too.
Only measured pixels belong on a surface that scrolls. The history now lives
on its own offscreen canvas that holds nothing but spectrum, and the visible
canvas is repainted from scratch each frame — history first, annotations
over the top. Nothing drawn as an annotation can reach the scroll buffer by
construction, rather than by remembering to clear the right strip.
The layout was muddled in the same place and is fixed with it: the gutter
used to be reserved along the bottom while the labels were drawn down the
left, overlapping the plot. Labels now sit beside the axis they describe —
down the left when frequency runs vertically, with room for "12.5k", and
along the bottom when it runs horizontally.
Verified both orientations by fingerprinting the gutter and the plot two and
a half seconds apart: gutter identical, plot changed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
AB's meter appeared to halt for a day and a half. Nothing had halted: the
capture child had run continuously since the machine was restarted, never
stalled, never exited, and was delivering twenty frames a second. Every one
of them was digital silence, because Dante Via stopped passing audio after
the reboot — the Timers module's LTC listener, on channel 16 of the same
device, had been reporting no signal for exactly as long.
The module's part in that was reporting "ok" through all of it and drawing
a steady -83, which is the -200 floor plus the +117 calibration offset. A
plausible number on a wall is believed, and a green dot beside it says the
number is worth believing. That is worse than showing nothing.
So silence is now a state the module names. Below -150 dBFS unweighted the
input is not quiet, it is absent: a microphone in a silent room still reads
about -83, and even a muted converter dithers. Five seconds of that on every
channel and health() goes to error naming the device and how long it has
been dead, the tiles stop drawing a level and say "No audio" with how long,
and the status dot goes red.
Verified both directions, which is the half that matters: BlackHole with
nothing routed into it is caught in five seconds, and a real microphone in a
quiet room at -49 dBFS is never flagged.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The current item followed the live ProPresenter document by title alone,
and a sermon's document is named after the sermon — Waters names them
"YYMMDD Series Title - Part #" — never after its plan item. So the last
song stayed current through the whole message: North Attleboro on 20 Sept
spent 65 minutes on one song in two of its three services.
A live document that reads like the plan itself — its date code, series
title and part — now lands on the plan's message item: an item titled
Message, Sermon or Teaching, else the first item under such a header. A
date code naming another day is a firm no, so last week's sermon stays
unplaced. Position in the ProPresenter playlist breaks a tie between
equal names, only while the playlist mirrors the plan item for item, and
never places a document on its own.
The log now says what it never said: every feed drop and reconnect, which
item the message document will land on, what placed each item and how,
and a live document that is not in the plan.
matching.test.js: 27 checks built from both campuses' real 27 Sept plans
and ProPresenter playlists.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The Layouts menu row reads "Grid Size − 12 +": the label is the setting's
name, the readout is the bare column count, and the default/finer/coarser
caption under the label is gone. The buttons keep their tooltips.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Stepping the grid coarser pulled tiles past the right edge back in and
left them wherever that put them — on top of one another, which the
board never allows under a drag. Tiles now settle in reading order: one
that would land on a tile already settled takes the first free spot, the
same rule a newly added tile follows.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Twelve columns fit a booth monitor and nothing else well: on a 4K wall a
three-wide tile is nearly a thousand pixels, and on a small screen four tiles
across is the most the grid allows. A stepper in the Layouts menu now cuts
the board into 8, 10, 12, 16, 20 or 24 columns. Row height follows the
column width, so cells keep their shape and a 3×2 tile is still a 3×2 tile,
only smaller — finer lets tiles be smaller so more fit, coarser suits a
screen read close up.
It is this browser's choice, not the server's: two displays at one location
can want different densities, so it saves with the layout — the local one
and a named one on the server — and a named layout comes back at the density
it was arranged on, which the load message says. An older layout with no
density is the default twelve.
Tiles keep their grid units when the density changes; the board simply gets
denser or coarser around them, and a tile now past the right edge is pulled
in. A module's minSize is enforced in cells as declared, so on a finer grid a
tile can be smaller than it could at twelve, which is the point; a new tile's
defaultSize is scaled to the density so it lands at about the size its author
meant. Leaving fullscreen goes back to the chosen density rather than the
stylesheet's twelve — that reset predates any notion of a chosen density.
Shell 1.9.0. The round-trip test carries gridCols and drops a value off the
scale.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The module arrived in the previous commit exactly as it runs in North
Attleboro. This brings it to the interface rules every module follows and
puts it where the admin page can offer it.
The three hide-toggles — names, mic labels, parallax — sat in a folded
"Show / hide" group. Rule 8 puts four or fewer at the top of the gear with
no group at all, so they are ungrouped and first; the Birthday group keeps
its switch. The test-mode select loses its help line (rule 5). Everything
else in the gear was already within the rules, and every title-bar control
was already an icon with a tooltip.
The Planning Center mock moves out of the module folder to tools/, beside
the other upstream mocks, so an install does not download a development
tool with the module; README and CLAUDE.md say how to run it. The version
goes to 1.11.0 so the copy already installed on NA is offered the update
from the repo, and the module joins the README's list.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
The backstage stage board as a ProdDash tile: everyone scheduled on the
band, vocal and production teams for the upcoming Planning Center plan,
placed into their mic slot with photo, name and mic/pack label. Birthdays
in the service week get confetti. The display half of the standalone Band
Lineup Electron app, ported to a module — imported here exactly as
installed on NA-Ingest (v1.10.1), so the repo is where it lives from now on.
The bug behind this: for seventeen days the server rebuilt each saved tile
from the fields it knew and dropped the rest — including `variant`, so a
layout of five different timers came back from the server as five "All
timers". The client had carried `variant` since multi-tile modules arrived;
the server was the one that lagged, and it is the part of ProdDash that
always will (a booth updates weeks after the dev machine). So the server
stops deciding which tile fields exist: it checks the shape of the ones it
knows, keeps any other small plain-JSON field a tile carries, drops only
retired ones, stamps the file `v: 2`, and refuses a layout over a hundred
tiles instead of quietly saving the first hundred. A test round-trips a
fully-populated tile plus one the server has never heard of, and fails if
anything is lost.
Then the rule the arrangement is loaded by. A layout keeps every tile it is
given; the board shows the ones whose module is installed and enabled and —
for a tile added from a picker entry — whose entry is offered right now. The
rest keep their place and draw nothing, and appear the moment the registry
offers them again: a Service timer waits for PCO Plan, a transcript for its
channel. Never the module's default in their place. Because a module's entry
list can change with no event to mark it (PCO Plan comes back and, seconds
later, the timers module is offering its timers again), the registry is
re-read every few seconds while anything is folded away, and not at all
otherwise. The timers module's placeholder card for a missing solo timer
goes with this: nothing is drawn, the dot says why.
Loading says what it did — "Loaded 'FOH': 9 tiles, 2 not shown — Service,
#ops aren't offered right now" — with Undo in the message. Saving over a
name asks first, and the server keeps the copy it replaced one back; Admin →
Saved layouts gets Restore previous, which swaps the two so a restore is one
restore from undone. The browser keeps one good copy of its own layout
behind the live one and falls back to it when the live one will not parse.
Shell 1.8.0, timers 1.4.1. The layouts already saved on the booth machines
had their variants stripped before this and have to be saved again from a
browser that still holds the right tiles — nothing here recovers what the
files no longer contain.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Two things the audit found, both about the archive path every shipped .app
uses.
**Settings could be wiped.** The update kept config/modules.json and
config/layouts/ by name, but not config/proddash.json — which holds the admin
PIN, the theme and the forge pointer. On an install whose data directory IS
the app's own config/ (an unwritable Application Support, or PRODDASH_DATA_DIR
aimed inside the app), an update wrote the repo's shipped defaults over it:
the theme went, and admin silently unlocked. Nothing inside the data
directory is written now, wherever it happens to sit, and the two legacy
names stay listed for an install whose state never moved out.
**A bad update was unrecoverable.** No rollback, and a fresh .app would not
help: seeding only replaces an *older* copy, and a broken one carries a
higher version — so the fix was deleting a folder nothing told you about.
Three changes, in the order they catch things:
- The download must parse before anything is swapped in. `node --check`
against the new server.js, and a failure is reported with the syntax error
and leaves the running copy untouched.
- Past that, the copy being replaced is set aside as `app.previous`, and
dropped once the new version has stayed up ninety seconds — so that folder
existing means "the last update has not proven itself yet", and never
anything older.
- The launcher, which already gives up after four stops in two minutes, now
looks for that folder before it does and puts the old copy back. Once: if
the old one will not run either, the update was not the problem.
The launcher half only reaches a machine with a new .app; the server half
travels by the normal update.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Deploying 0.3.0 to AB crashed the ProdDash server on restart:
TypeError: Cannot read properties of null (reading 'streams')
at server.js:273 req.on('close', () => state.streams.delete(res))
teardown() nulls `state` and then ends every open stream, which fires each
request's close handler — and the handler reached back through the `state`
it had just discarded. Any re-init with a browser connected would do it, so
saving admin config while a dashboard was open was enough to stop the whole
server, every module with it.
The stream handler now closes over this mount's Set rather than the module's
state: after teardown it deletes from a set nobody reads, which is harmless.
The same hazard existed anywhere a callback could outlive the mount — the
keepalive ping, a frame arriving from the listener mid-teardown, and the
/measure interval — and each is closed over or guarded now.
Reproduced first (open an SSE client, save config, watch it fall over) and
confirmed fixed against the same sequence.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Rebasing onto main landed rules 5 to 11, which this module predates and
broke in three places.
Rule 5 — no explanatory text in a tile's gear: the Banding and Smoothing
fields each carried a paragraph. The Banding one said something true and
worth knowing, that below roughly 200 Hz the window length limits real
resolution however fine the setting, so it moves to the source where the
band plan is built rather than being lost. Option labels shortened with it.
Rule 11 — one control per action: the spectrogram offered its axis both as
a gear field and as a title-bar button. The button stays, since that is
what was asked for, and the field is gone. The setting still persists with
the tile; saveInstanceSettings keeps keys that no schema declares.
Rule 9 — title-bar controls are icons: the axis button was drawing a "⇆"
as a label. It is an inline SVG now, with the tooltip carrying the words
and naming the current direction rather than the action.
Rule 8 also asks that a tile can switch off what it draws, so the readout
gained a Caption toggle and both graphs gained Labels.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The number was capped at 6rem, so a readout given half a wall stopped
growing at the size it reached in a small tile.
Removing the cap was not enough on its own. Sizing text off container units
means choosing a coefficient for some assumed number of characters, and the
value here is not a fixed width: the coefficient that fills a tile nicely
for "-45.9" pushes "-100.5" nine percent out through the side. Measured it
rather than guessed — the tile now sets a trial size, reads the width the
actual text wants, and scales to fit. That runs only when the tile size or
the length of the reading changes, not on every frame.
It now fills about 98% of the width wherever width is the limit and 92% of
the height wherever height is, at every tile size from 180x70 up to
1400x700, with no overflow on either axis. The CSS keeps a conservative
size for the first paint before the measurement has run.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Every control asked for here is a view choice, so none of them may reach
the capture child — two dashboards watching the same room must be able to
disagree about how to draw it without disturbing the measurement or each
other. That set the shape of the change: the child now sends everything
anyone might ask for, once, and each tile folds it down.
Averaging (was "Reading") is Fast, Slow, Inf and None, and the child
computes and sends all four every frame. Fast and Slow are the IEC time
constants, Inf is Leq since reset, None is the block with no time constant
at all. The admin-wide time constant is gone — it was never a property of
the room.
Banding is 1/3 through 1/48 octave. The wire always carries 1/48, and
because every banding offered divides 48 a coarser band is an exact sum of
finer ones in the power domain. Nothing is resampled and no tile's choice
is visible to another. The transform grew to 16384 points to have bins fine
enough to mean it; below roughly 200 Hz a 1/48 band is still narrower than
one bin, which the Banding help says out loud rather than pretending.
Each tile kind now carries its own settings, which is the fix for the
complaint that started this: the RTA was offering an alarm threshold and a
metric picker that did nothing to it. An entry's instanceSchema replaces the
module's, so a gear popover only holds controls that change that tile.
Also: frequency labels in three tiers that thin out as a tile narrows
instead of smearing, gridlines off by default on both graphs, RTA smoothing
as a damping factor on the drawn value only, a title-bar toggle to swap the
spectrogram's axes, and the threshold blurb removed.
Settings saved by 0.1.0 are mapped forward — a layout on a wall should not
start reading dashes because a field was renamed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The missing-binary fallback was a spawnSync, copied from the LTC listener
where the source is a third the size and builds in a few seconds. On the
production Mac this file took nearly two minutes, and spawnSync held the
Node event loop for every second of it: one module's optional helper froze
the entire dashboard, every other module with it, and then did it again
thirty seconds later on retry.
The build is now a plain spawn that reports "building" and calls start()
itself when it lands, so nothing waits on it. A failed build says so once
instead of spinning, and a build that never finishes is given up on after
five minutes.
The real lesson is the one in the comment: the binary should already be
there. It is built beside the app, or copied from any Mac of the same
architecture. Building on the machine that might be running a service is
the fallback, not the plan.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
michael
christian