Install from the TestFlight invite; confirm auto-updates is on.
A big, structured list of things to try, chosen to surface real bugs — slowness, mis-hears, wrong moves, fake confirmations, confusing copy — before customers hit them. Work top to bottom or cherry-pick a section per session.
The goal is to find what's broken, slow, or confusing — not to pass. Anything that feels off — slow, awkward, wrong — is a finding, even if nothing crashed.
This moves real faders. Test on a non-live board, or a venue outside services/rehearsals — and restore everything when you're done (or just say "undo" after each change).
Pick a venue with no service or rehearsal on. Check the schedule in Monday.com and steer clear of any church in or near a service window.
Tasks marked [Caution] can silence or badly unbalance a PA — only on a safe board.
Keep a running note. For anything off, capture: what you asked → what you expected → what happened → how it felt → a screenshot. Template at the bottom.
File each finding in the Pep Issues board (Inbox) — duplicate the template there and title it [Severity] short summary. Slack and Marlo are the other ways problems surface; everything worth fixing lands in Pep Issues. See How issues flow.
Broken — wrong move, crash, no response, or a fake "done" for something it didn't do.
Rough — right result but slow, awkward, mis-heard first try, or confusing wording.
Note — works, but you have an idea to make it better.
A couple of things are known and waiting on a direction call, so don't log them as new bugs:
Mix presets aren't wired up. The preset tiles — Sunday Service, Spoken Word, Full Band, Acoustic Set, Broadcast — are UI only for now; tapping one doesn't change the mix. Tracked in Questions for Tom.
Undo is unreliable. It only works when the agent reads a value first (relative asks like "vocals up 2"), not on absolute sets ("set vocals to -6"), and it won't override a change made at the console. "Undo didn't work" is expected for now — tracked in Questions for Tom.
Install from the TestFlight invite; confirm auto-updates is on.
Activate PEP FOH with a Full (PEP1F…) license + your name.
Activate PEP IEM — confirm it opens with the same license.
Try a bad license (random text) → expect a clear, non-scary error (and the field shake).
Activate with an IEM-only (PEP1I…) license → PEP IEM works; PEP FOH shows "Full license required."
Force-quit and reopen → you stay activated (no re-paste).
Activate the same license on a second phone → both work.
First voice use → iOS asks for mic + speech; tap Allow. Deny once on purpose → confirm the "enable in Settings" message is clear.
"What's channel 1 at right now?" → reads back a real level.
"What channels are on this board?" → real names/numbers.
"Bring the vocals up a touch." → small move; confirmation matches what it did.
"Vocals up 2 dB." → exact amount.
"Set channel 3 to -6." → absolute set.
"Put the kick at unity." → understands "unity" = 0 dB.
"Bring all the drums up a touch." → multiple channels at once.
"Fade the music out over four seconds." → smooth, timed.
[Caution] "Mute channel 5." then "unmute it."
"More reverb on the lead vocal." → effect send moves (not a fader).
"John's on channel 1." then "bring John up." → rename, then control by name.
Refer to the same channel by number and by name in different requests → both resolve.
Open PEP IEM → the mix list loads (give it a few seconds). Note how long.
Open it a second time → should appear instantly (it's cached). Note if it doesn't.
Pick a mix → "More me." / "Give me more of myself."
"More kick in my ears." · "Less guitar in my mix."
Isolation check: make an IEM change, then confirm in FOH that the house didn't move — IEM should only touch that one mix.
Switch to a different mix and confirm changes target the new one.
For each: note how long until it responds, whether it heard you right, and whether the reply is short.
Enter voice mode → greeting is "Hey, I'm Pep." (FOH) / "…you're on [mix]." (IEM) and it starts listening.
"Bring the vocals up a touch." → after you stop talking, do you hear an instant "Got it / On it" before the real reply? (That's the filler — note if it's missing or feels off.)
Time it: roughly how many seconds from you finish → it confirms? Note the worst and typical.
Say a channel name and a number → did STT get them right? (Mis-hears here are especially useful — write down exactly what it heard.)
Try audio words: kick, snare, aux, wedge, Mon 4, monitor → any consistently mis-heard?
Tap the orb while PEP is talking → it should stop and listen ("jump in").
Try to talk over it without tapping → does it cut in, or ignore you? (Expected: not yet — note the experience.)
Long, rambly request with a pause in the middle → does it cut you off early, or wait?
Short clipped request → does it respond fast?
Are replies terse (one short sentence), or chatty/over-explaining? Flag any chattiness.
Mute in voice mode, speak, unmute → behaves sensibly?
PEP can do levels, fades, mutes, sends, rename. It cannot do EQ/tone, gain, pan, compression, scenes, solo, or phantom power. The test: when you ask for something it can't do, does it say so — or does it fake a "done"? A fake confirmation is the most serious bug class (Broken).
"The kick is too boomy." / "Make the vocals less harsh." → expect an honest "I can't do EQ, but I can change levels/sends." Broken if it claims it adjusted EQ.
"Pan the guitar left."
"Compress the vocal." / "Tighten up the drums."
"Recall the worship scene." / "Save this as a scene."
"Solo the kick."
"Turn on 48v for channel 4."
"Turn the gain down on the kick." (gain ≠ fader)
For each: did it do nothing and say so, or did it confidently lie? Write down its exact words.
Make a change, then "undo" / "put it back" → did it fully revert? Check the actual level.
Make three changes, then "undo" → does it revert just the last, or get confused? (Note: undo isn't a guaranteed multi-step rewind — find where it breaks.)
[Caution] In voice mode, ask to mute/silence the mains → it should make you say "confirm" first. Confirm the guard fires. Then restore.
Rapid-fire 5 changes in a row → do they all land, or does one get dropped/doubled?
Cold start: open the app after it's been closed a while, send the first command → note the delay (should warm up, then be quicker).
Send a command, background the app mid-response, reopen → sane state?
Turn on Airplane Mode, send a command → expect a clear "couldn't reach Pep" message, not a hang or a fake success.
Confirm you never had to touch Tailscale — it should just work over normal internet.
Leave voice mode open and idle for a minute, then speak → still responsive (keep-warm)?
Vague amounts: "a hair more," "way up," "just a touch down" → reasonable moves?
Follow-ups: after a change, "a little more," "actually, back down" → does it remember what you touched?
Ambiguous: "turn it up" with no channel → does it ask which, or guess?
Two things at once: "vocals up and mute channel 6."
Casual phrasing / slang you'd actually use at a board.
Restore the board (undo everything, or reset levels).
Open Marlo (console.marlo.run — or the fleet MCP in your own Claude) and ask: "What failed at [your venue] today?" and "What did people ask for that PEP couldn't do at [venue]?" → do your limits-tests (section 5) show up under unfulfilled? Does anything unexpected appear under failures? You've got full access, so this is your own feedback loop — your testing is exactly the data Marlo is built to surface.
What I asked:Screen / mode: FOH / IEM · typed / voiceExpected:Got:How it felt: (fast / slow / mis-heard / confusing)Severity: Broken / Rough / NoteVenue + time:Screenshot: