A new fix kind "companion_press" presses any Companion button over its HTTP API (page, row, col).
A fix marked "failover" runs by itself within seconds of its check failing: inside the "failover"
windows (Sunday 06:30-13:30), even while ALARMS is muted, once per fault, never back the other way.
The hidden STB AUDIO key (inside AUDIO) watches Studio B's ping every second, Pro Tools over SSH
every 2 s and its Dante device; each check's fix is the FOH Backup press. The keys say ON FOH BACKUP
until STREAM STB on the Restart page switches back. Settings page: the press builder, the failover
tick and undo, the Failover calendar, help entry "A fallback button".
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
- DIGICO (in the AUDIO group): Companion's DiGiCo_FOH connection, the digico-relay helper on
NA Ingest (hold AUDIO restarts it), and a ping to the console at 10.1.60.100 (the OSC
connection reads good even with the console off).
- Settings page: "+ Key" on an unwatched Companion connection asks which group key to put it
in when the Status page has no free spot (it used to make a key nobody could see); the
connection field of a check suggests Companion's connection names.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- readiness.py looks every 2 minutes (never while a HyperDeck records) at this weekend's PCO plan
and what is in place for it: the Ableton set (songs, mix, routing, lighting cues), ProPresenter
(every PCO item, MC announcements, baptism names, timelines, backgrounds), the LTC note of every
song against the timecode the tracks play, the Band Lineup TV, Resolume's LED cross, and the
tracks and ProP bundles for the locations.
- READY key on the Status page (row 2, col 7; SPOTIFY now shows inside SHOW) opens page 24.
Amber until done, blinking red from Saturday noon; not part of ALL. Tap keys for BOH mix,
mics and stage reset every Monday.
- Holds fix what can be fixed: re-sync the lineup, put the cross back up, write LTC into PCO,
copy tracks to locations, look again. Hold READY fixes everything.
- Settings page: a READY page tab (live keys, drag to move, why each key is amber), and Apply
re-imports page 24.
- Auto-fix: "while_muted" fixes keep running with ALARMS muted (Spotify playlist).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Overview jobs: Change the playlist, Add or remove a HyperDeck, Add a MIDI link, Add a computer,
Make a PC program startable; 'Add a key for it' on unwatched Companion connections.
- Computers > Setup: steps with a [how] each and the set-up scripts generated per computer by
settings/scripts.py from the templates in remote-access/ (no site-shaped value left in them);
Remove this computer drops the fixes and checks that used it.
- Site setup > Site: the running and staged code, Install the staged folder (a one-shot launchd job
running install.sh --keep-config, refused while live), Export site bundle (zip), New campus from a
bundle (this campus's site block stays; every copied computer is flagged until touched).
- sites/: na -> the repo config, _template = the empty campus; install.sh --site <id>.
- Review reads appended checks as 'added'; tests for all of it.
- Deployed page-only to NA Ingest. (config.json, status_panel.py and readiness.py are left to the
READY-page work in progress in the same tree.)
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
settings/builders.py renders a fix's command from a builder (template) and its fields; validate
refuses a config whose stored command no longer equals what its fields build; the Fixes screen and
the remote checks in the key editor rebuild through /api/build and /api/build-check. config.json
adopts templates for all 27 fixes and 5 remote checks (ProPresenter's restart loop and Ableton's
quoting normalised, same behaviour); tests/test_builders.py (174 tests pass). Deployed to NA Ingest.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
- settings/apply.py: the Apply pipeline (check, keep history, write, restart the watchdog, re-import
pages by the marker-only rule, recolour page keys), rollback when the watchdog does not come back,
the recording guard with the Emergency (LIVE) override, history entries and Go back / Undo.
- settings/server.py: the write routes behind the editing code (draft PUT/DELETE with 409 on a
conflict, validate, apply, restore, undo, watchdog restart/stop/start, commands to the watchdog,
the switcher file upload / pick newest / discard, gear notes).
- settings/ui/app.js: every screen grows inputs with Editing on; keys (checks, fixes, hold-always,
members), computers, fixes, site setup (schedules, Dante devices, Spotify, ATEM), Review & Apply
with live progress, History with Go back and Undo, Run this fix now.
- schema.unmask, push_page.push, nav_keys.colour; tests for all of it (163 pass).
- Verified live on NA Ingest: a page-only Apply, a restart Apply, Undo, history.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
Overview (problems in the room with the hold action, the live deck, one line per computer),
Keys (both pages, what each key will say and what holding does), Computers (map with the keys
that turn red, cards with a light per thing that runs there from the watchdog's snapshot and a
read-only probe, Gear, Setup), Logs, History (the installer's copies for now), Help (four cards,
the generated key guide, cheat sheet and Where things are, glossary, Companion facts), Site setup
and Fixes read-only. The editing code is checked (with a lockout) but nothing can change yet.
Also: faults are attributed to the computers their check is about; the installer refuses while a
HyperDeck is recording (--now overrides); the Lighting PC's MIDI session is listed; the probes read
sudo's answer correctly and expand $HOME outside the quotes.
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
- schema.py: every setting with type, default, help, owner screen and the rules a config must
pass; tests/test_schema.py keeps it in step with the code (every key the code reads is declared,
every default matches, check types and fix kinds agree, the glossary covers every word a key can
show, today's config still builds today's pages apart from the key notes)
- status_panel.py: VERSION + started.json; SIGTERM lets a running fix finish and starts nothing new
(RESTARTING SOON); fixes-running.json and live-now.json for the settings page; state.json keeps
the ALARMS override, acknowledgements, auto-fix tries and fix times across restarts, keyed by
what a check is about rather than its position; 'test' and 'resurvey' commands; queued commands
dropped at start; overnight windows; a HyperDeck switched off is always switched back on;
the spotify fix kind (playlist set once, in config 'spotify'); Windows remote checks batched for
cmd.exe with conditional echoes; every fix's kind and host validated; never_auto; wording and
numbers from config (site block, timing, page_hold_ms, auto-fix keep_months); hyperdeck_probe
removed; --check and --exit-timeout
- make_page.py: numbers from config, markers on every key, nav templates per Companion version,
--out; push_page.py: a page is overwritten only when every key carries a marker (foreign keys
named by position), secret-free fetch, host from the site; nav_keys.py: host from the site;
netgear.py: switches and the shared-port limit from config
- install.sh: validates with the new code first, keeps prev/ for --rollback, --keep-config,
derived ExitTimeOut, wider deploy list, waits for the start stamp
- config.json: site, timing, spotify sections; Windows hosts marked; deployed to NA Ingest and
both pages re-imported once (--force) so every key carries its marker
Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>
A LaunchAgent helper on that Mac (Accessibility) reads Options > Solo Mode
every second into a status file and ticks AFL/Latch on request. APPS
alarms PT SOLO NOT AFL (hold fixes it; in the Sunday auto-fix list) and
PT SOLO CHECK OFF when the helper is not running or not allowed.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Only files named exactly '<YYYY-MM-DD> <Weekday> Auto-Fix.log' in that
folder; checked when a new day's file starts and when the watchdog starts.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
~/Desktop/Auto-Fix Logs/<YYYY-MM-DD Weekday> Auto-Fix.log, written line by
line as it happens: auto-fix on/off, each problem and what was done (try n
of 2), the result, when the fault cleared, give-ups, and when it held back
(network problem, several Dante devices gone). Tests write to a temp
folder.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sundays 06:30-13:30, with ALARMS on, a fault that has lasted 20 s is fixed
by itself if the fix is gentle (reconnects, HyperDeck off/on, Tascam, MIDI,
rtpMIDI, starting a closed Vista/Resolume/OBS/Ableton/Translate, DVS on one
Mac). At most 2 tries 90 s apart; key shows AUTO-FIXING. Anything that
closes a running program stays a hold (load_config refuses it).
A review before going live (2 lenses, each finding verified) confirmed and
this fixes: a Dante-wide "gone" (Ingest's view) would restart DVS on every
Mac every minute - several devices gone at once now means hands off; the
try limit reset whenever a fix made its fault blink - tries are kept until
the fault has been gone 10 min; DVS on Ingest kills ProdCom - removed from
the auto list; restarting dead connections looked like a network burst -
only a connection good in the last 30 s counts; a restart key could kill
an app auto-fix was starting; concurrent MIDI reconnects overwrote each
other - now one worker. Auto-fix errors are caught so the panel keeps
running. 105 tests.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
After a deck restarts, Companion's HyperDeck connection does not come back
by itself and a plain restart of the connection does not help; switching
it off and on does. Each HD key's hold now does exactly that (even when it
looks green), waits 15 s and checks: DONE, STILL OFFLINE (deck pings) or
NO NETWORK. Decks are no longer rebooted by the panel; a recording deck is
left alone. Tested live on HD2 (off/on, reconnected in 17 s).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- A Restart page key flashes red "HOLD TO FIX" while any check that its
fix (or a fix/connection in its "watch") covers is down, e.g. DVS
TRACKS when DANTE misses NA-Tracks.
- Ableton's start and restart reopen the set Live last opened (read from
its Log.txt before quitting; templates skipped) instead of the default
template. Tested live: ABLETON and TRANSLATE restart keys.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- The remembered network blip is a flashing amber warning (new "notice"
level, not counted as DOWN): press NET to stop the flashing, hold NET 2 s
to clear it, otherwise it goes by itself after 5 minutes (was 3, solid).
- The user moved the Restart page to page 20: config, the RESTART PAGE key
on page 19 and both pages re-pushed. A scan of all pages finds no other
button pointing at the old page.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Page 23 "Restart": hold a key 2 s to restart DVS (4 Macs), ProPresenter,
Ableton, Church Translate, ProdCom, ProdDash, captions, the bridge or the
Tascam connection even when the status page says it is fine ("force",
shared 60 s guard). RESTART PAGE key on page 19, STATUS PAGE back.
- Holds carry Companion's clock; ones queued while the watchdog was down
are dropped instead of firing together later.
- TASCAM hold always reconnects and checks again after 20 s: HUNG - POWER
CYCLE IT (pings, control port hung), NO NETWORK, or turns a disabled
connection on. Tascam pinged at its Companion address.
- INGEST: ProdCom (:24480), ProdDash's page (:24500), the caption screen;
DVS on Ingest restarts ProdCom afterwards. RESI: the encoder (10.1.10.202).
- Windows starts check the app really started (FIX FAILED otherwise); OBS's
start task runs elevated (OBS is "Run as administrator"); setup script
does that for any such app and gives sshd restart-on-failure.
- push_page.py checks what is on the target page, not its name (pages 22
and 97 are named "PAGE" but hold buttons); --which restart.
- Companion custom variables are written in the request body (no URL limit).
Reviewed by a 3-lens review with per-finding verification; 7 confirmed
issues fixed. Tested live: OBS hold opens OBS; Restart page TASCAM key.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
It flickers to "warning" every 15-20 minutes and made APPS blink for
nothing. "ignore_warning" on a conn check keeps offline, error and
disabled as alarms.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A power-off test today took 18-27 s to show on RF (Companion's Shure
modules' own timeout). Each Axient, ULXD and PSM receiver is now pinged
every 3 s: "AXIENT 1 NO PING" in about 6-9 s. A ping check may name a
Companion "conn"; its address is read from Companion's settings database
(read-only, every 60 s), with "host" as the fallback.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The Art-Net heartbeat moves from APPS to a VISTA tile ("VISTA / SILENT";
"VISTA SILENT" on SHOW and ALL), hold starts Vista if it is not running.
SHOW now groups MIDI, APPS and VISTA. Tested live before the move: closing
Vista and Resolume alarmed, and each hold brought them back (Vista 5 s,
Arena 9 s after the hold).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- Hold MIDI on "LIGHT PC MIDI GONE" starts the rtpMIDI service, then its tray
app, only if they are not running (a restart drops every session).
- The Lighting PC's rtpMIDI service now has Windows recovery actions
(restart after 5 s, 5 s, 30 s); windows-setup.ps1 sets the same.
- coremidi.disconnect_peer (Disconnect in Audio MIDI Setup), used to test the
alarms: all six links dropped and came back (four by themselves, two by
the hold).
- Audio MIDI Setup need not be open: quitting it on NA Ingest left the
session connected.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Adds Tracks -> Livebox, Tracks -> Studio B and ProPresenter -> Lighting
(read on the ProPresenter Mac over SSH), each with the hold-to-reconnect fix.
coremidi.py now recognises the network MIDI driver on macOS 26.2 too
("AppleMIDIRTPDriver"; 26.5 calls it "AppleMIDINetworkDriver"), which is why
the ProPresenter Mac's session looked empty before.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
macOS's MIDINetworkSession API only reaches the default session (blank and
disabled on NA Ingest and NA Tracks), so coremidi.connect_peer writes the
named session's "apple.midirtp.session" settings with the peer added, as
Audio MIDI Setup's Connect does. Existing peers are passed back unchanged.
The watchdog remembers each peer's last address (config "address" is the
fallback) and runs it locally for NA Ingest's session or over SSH on the
Tracks Mac for NA Tracks' (FX lights). Checked live: both Macs answer
"already" while the Lighting PC is connected.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Tracks sends the FX lighting cues to the Lighting PC through NA Tracks'
network MIDI session. A midi_peer check may now name a "host": the watchdog
reads that Mac's sessions over SSH every 5 s, sending this folder's
coremidi.py over stdin (nothing installed there). Unreachable = no verdict;
the Tracks Mac's own ping check covers that.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Its Bonjour firewall rules only allow Public; switched to Private it dropped
out of the MIDI directory and could not rejoin NA Ingest's session.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Tested live: restarting rtpMIDIService brought the Lighting PC's Bonjour
advert back but dropped it from the NA Ingest and NA Tracks sessions, and it
did not rejoin them. macOS only exposes the default network session to
scripts, not NA Ingest's named one, so there is no safe automatic reconnect.
MIDI now has no hold fix; the README says how to reconnect by hand.
ssh fixes run with LogLevel=ERROR so warnings stay out of the log.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- OBS: start OBS on the LiveStream PC if it is not running, then reconnect
Companion. If the PC does not answer, send Wake-on-LAN and show WAKING PC.
- ARENA: start Resolume Arena on the Lighting PC if it is not running, then
reconnect Companion. VISTA SILENT: start Vista if it is not running.
A running OBS, Arena or Vista is never closed.
- MIDI (LIGHT PC): restart the rtpMIDI service; it reconnects to its saved
peers (NA Ingest, NA ProP, NA Tracks).
- Fixes may set "reconnect" (restart the check's Companion connection after
the command succeeds); hosts may set "mac"/"wake_broadcast" for Wake-on-LAN.
- windows-setup.ps1 finds OBS on any drive and Vista, and leaves Wake-on-LAN
alone when it is already on (changing it can reset the adapter).
- Tests stub the fix executors so a hold in a test never reaches a machine.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
SSH to media@10.6.7.13 (Windows OpenSSH, cmd.exe): kill a hung Bible App,
then run the StatusPanel\Start BibleApp task, which starts it on the
logged-in desktop.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
- The hold-to-fix deck probe takes each HyperDeck's IP from its Companion
connection's "ip" variable, so the config no longer repeats them.
- APPS shows "VISTA SILENT" when the Lighting PC's Art-Net poll (every 0.5 s)
is not heard for 5 s: a crashed or closed Vista shows in seconds, not the
minute macOS takes to drop a network MIDI peer.
- tascam-quiet/ builds the Tascam DA-6400 module 1.3.1-quiet: the recorder's
"Illegal Status" warning is logged once a minute, not 5 times a second.
- README: hold-to-fix table fixed and updated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Every 10 minutes the watchdog maps each Dante device: name and IP from
Companion, MAC from NA Ingest's ARP table, switch port and PoE state from
the Netgear switches in dante_switches, saved to dante-map.json. A DANTE hold
on a missing PoE-powered device power-cycles the port it is on now (found
live by MAC, or its last known port if it has aged out and nothing else is
there); never a port with more than 2 devices. Re-patching, adding or
renaming devices needs no config change. First live map: 19 devices, the 4
Studio B boxes on PoE.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
STUDIO-B-SOLO-MON, MON, TB and TB-MON map to ports 0/14-0/17 on the Netgear
M4250 Dante switch 10.66.1.102. netgear.py poe_cycle first checks the port's
MAC table and refuses if a different device is plugged in there, then runs
poe reset. Tested 2026-10-08 13:10 on SOLO-MON: dropped off Dante and was
back about 17 s later. install.sh now copies netgear.py too.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fixes come from config.json only and run over SSH with NA Ingest's
status-panel key (ProPresenter restart, Ableton / Church Translate start,
reopen DVS), locally (Resi bridge, captions, ProdDash) or as a HyperDeck
reboot. A green button does nothing, each fix waits 60 s before it can run
again, and a recording deck is never rebooted. New remote checks ask the Macs
every 10 s whether their apps are running. Group buttons and INGEST now have a
hold too; tests never run a real fix or open an SSH connection.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Ping failures now read NO PING, Companion connection errors OFFLINE, and a
watched connection that is disabled in Companion shows amber DISABLED instead
of going quietly grey.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The UHD Jr was switched off and the NovaStar module still reported
Connected (even after a restart), so LED stayed green. LED now also checks
that each processor answers ping; the Resi decoder (send-only RossTalk) gets
the same.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The converted key's "startup" page resolves to the current page on surfaces
set to "use last page", so a tap did nothing. The tap page is now
nav_tap_page in config.json (1).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
A tap still goes to the start page, now on release so a hold does not flash
through page 1 first. Hold time is nav_hold_ms in config.json.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
nav_templates.json holds Companion's own UP / page number / DOWN keys after
'convert to regular button'; make_page.py colours them from the ALL tile and
nav_keys.py copies them onto all 96 other pages (288 keys, recorded in
nav_locations.json; --revert restores the built-in keys). Also rewrites the
README's top half, which an earlier edit had duplicated.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
It uploads the page file over Companion's own API (tRPC over WebSocket, the
same calls the web UI makes) and imports it into page 19, refusing unless that
page is empty or already the status page.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
DANTE now alarms for all 19 devices by name. MIDI covers the Lighting PC and
SuperRack peers plus the Lighting PC and NA Tracks sessions; APPS covers
ProPresenter, Ableton, the tracks Mac and the Bible TV app. Five buttons
became three, and a check's tag names the device in the message.
dns-sd pads hours before 10 AM with a space, which the Bonjour parser did not
allow for, so every session read as gone each morning. Parsing is now its
own function with a test for both cases.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
It now reads 19/19 to match Dante Controller. NA-Tracks, SuperRack and the
ProPresenter Mac still alarm on their own buttons, so a drop is not flagged
twice; DANTE just shows 18/19 while one of them is missing.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
All 19 Dante devices are now watched: 16 on DANTE, and NA-Tracks, SuperRack
and the ProPresenter Mac on their own buttons so a drop is not counted twice.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
status_panel.py runs on NA Ingest as a LaunchAgent. Every second it reads
Companion's connection statuses and module variables, adds what Companion
cannot see (network MIDI peers, Bonjour sessions, pings, launchd services,
disk, Resi cloud, the Bible TV app), and writes one JSON custom variable that
page 19 renders. Buttons acknowledge alarms, mute them, and restart a stuck
Companion connection on a two second hold. If the watchdog stops, the page
turns purple instead of freezing on stale green.
make_page.py builds the page export from config.json; tests/ runs offline
against a fake Companion and a CoreMIDI loopback.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
