Seven Thousand Errors, One Fader, and No Date

Short version: the beta is being built, hard, and I am not going to give you a date I cannot keep. I gave you one already. It was May. May came and went.

So instead of a new date, here is what actually happened — including the part where I found out the studio had been throwing seven thousand errors a second and still looked completely fine.

The fader that was lying to me

The console’s faders moved. The channels responded. Everything looked right — and the mix review screen sat perfectly still through every single take.

The chain runs from the fader mesh, through the slider’s progress notification, out as an interaction event that each of the 56 channel strips picks up. One link in that chain was not firing. Now it does, and you can watch it happen.

The LEVELMATCH MIX REVIEW screen at two seconds — channels 1 to 24 ramping up, channels 25 to 48 decaying — Left: two seconds in — the top bank ramping up, the bottom bank decaying. Right: nine seconds in — the phase-shifted sine has become an arc across the channels. In every render before the fix, both of these screens were frozen.

The LEVELMATCH MIX REVIEW screen at nine seconds — the phase-shifted sine drawn as an arc across the channels — Left: two seconds in — the top bank ramping up, the bottom bank decaying. Right: nine seconds in — the phase-shifted sine has become an arc across the channels. In every render before the fix, both of these screens were frozen.

That is the difference between the fader moves and the fader does something. It is not a difference you can see in a screenshot of the desk, which is exactly why it survived as long as it did. You have to look at what the signal is doing, and until the mix review screen worked, nothing was looking.

The same sine, seen from the room: 56 faders, each one phase-shifted a little further than the last. Fifteen seconds, no audio.

7,519 errors per session

While chasing that, I turned on verbose Blueprint logging and found the studio generating 7,519 runtime errors every play session. Not crashes. Just constant, silent complaining that nobody was reading.

The cause: the laser pointer components were being destroyed at runtime, but the character kept holding references to the corpses and ticking them anyway. Two validity gates, and the parent properly clearing its variables after the destroy, and it is zero. Not five hundred. Zero.

Nothing on screen looked different afterwards. But everything underneath got quieter, and quiet is what you want before you invite people in.

Why there were so many at once

Here is the honest version. For a long stretch, I could not run this build in VR at all.

There is an older, parallel version where VR worked and kept working. This one went through refactor after refactor — the console, the patchbay, the tape system — and through all of it the VR path was not something I could put a headset on and check. It still compiled. It just was not being run.

This is what a refactor touches. Every one of those components has an interaction path attached to it, and none of that path complains when it quietly stops being correct.

Code that is not run does not stay correct. It rots quietly. Every refactor that touched a shared component left the VR side a little more wrong, and nothing objected, because nothing was looking. None of these were mistakes made in one bad afternoon — they accumulated across months of work that was, from VR’s point of view, invisible.

Retro Recordings XR in VR: a disembodied gloved hand lying on the studio floor while a second ghost hand hovers over the console faders — A hand that renders as a ghost while it grips. A hand left lying on the floor. Every popup on the desk stuck open at once. A control panel whose labels have walked into each other. None of this is exotic — it is just what an interaction layer looks like after months without anyone in it.

So that is the phase I am in: paying that bill. All of it has to be fixed before I can get back to what I actually want to be doing — the rest of the console’s functionality, the tape system, the multitrack. Not because the VR bugs are more interesting, but because there is no sense in building features on top of an interaction layer that is broken in ways nobody has looked at.

What the headset keeps teaching me

Two VR test rounds this month, and both found the same category of problem: work that was invisibly wrong rather than unfinished.

The onboarding menu rendered perfectly on the monitor and materialised inside my eye in the headset. It was anchored to the camera manager, which does not follow the headset in VR. It is now world-anchored, with a height clamp so it can never sink below the console line.
The knobs stopped turning entirely. Empty overrides in a subclass were silently eating the native rotation code underneath. Restored, with the haptics back.
My hands were rotated about 105 degrees, because a hand-tracking controller profile gets loaded that I was not expecting.

Menu geometry is its own genre of wrong. Here it is during development, rendered at a scale that puts three words across the whole console and hides the room behind them.

None of these show up on a desktop test. All of them end a session in VR. This is the thing that makes scheduling dishonest: the remaining work is verification, and verification keeps producing new work.

The rack

Next to the desk stands the outboard rack, and it is what I am working on right now. Eight units — my own builds, inspired by a Fairchild 670, an LA-2A, an 1176, a Pultec, a Distressor, a Lexicon 224, a Space Echo and an H3000. Clones, not reproductions: they do not carry anyone else’s name, and they are not trying to be the originals. The originals are documented in the archive; these are what I built after reading them.

Routing and patching are in place — real cables you pull out of the bay and push into a port with your hands, not a menu. What is happening now is the front panels: making the controls actually do what the panel says they do, one unit at a time. The optical tube compressor is first. The rest follow the same recipe once that recipe is proven.

The insert send and return ports, one pair per channel. Each one is its own actor with a socket, a ghost jack and a detection sphere — which is how a cable knows it has found a home.

I have written up all eight, and the twenty-four other boxes that shaped recorded music, in the outboard archive — designers, mechanisms, and why studios still hunt for the originals.

So where is the beta?

Being built. Genuinely, daily.

The onboarding flow — walk in, pick Visitor or Engineer, and get taken through teleporting, turning a knob, moving a fader, patching a cable, loading a tape and playing it back — is code-complete and compiles clean. What it has not had yet is a single human running all six steps start to finish in a headset without hitting a wall. That is the last gate, and it is the one I cannot put in a calendar, because every test round so far has surfaced something that only exists in VR.

And here it is behaving: four buttons, sitting where they belong, with the mix review screen glowing away on the left. This is the first thing you will see when you walk in.

I said May, and I am not going to do that to you twice. So: no date. What I can tell you is that what is left is verification, not construction — and the moment a real person can walk the whole flow without me apologising for something, the beta goes out.

Subscribers and the Discord hear it first, in that order.