Skipped, pre-confirmed
In the released app — 2.1, the build in listeners' hands — there is a diagnostics line, deep in the advanced pages, labelled "SET_CUR OK". SET_CUR is USB audio vocabulary, not the Engine's invention: it is the command a host sends a DAC to set, among other things, the sample rate of its clock. Most of the time the line reports plainly whether that command succeeded. But under the right conditions it reports something more interesting: "skipped (pre-confirmed)" — meaning, this session did not need to send its own rate command, because an earlier write in the same transition already confirmed the rate; the redundant one was skipped.
That claim is false, and it has been false since the day it shipped.
A reader can verify the line exists — it is on the glass, in the released build, and it appears in the diagnostics export the app will generate on request. What a reader cannot verify is its truth; the rest of this entry rests on reading the source, and I will say so plainly: these are source claims, not checkable from the app.
Here is what the source says. The skip the line testifies about was designed, reviewed, and plumbed — a flag threaded through every factory that builds a playback session, consulted at exactly the right moment, ready to elide the second rate write when the first had already been confirmed. And then it was never switched on. Every production site that could raise the flag passes a literal no. All six of them. The session always issues its own rate command, on every transition, and always has. The mechanism this diagnostic line reports on has never fired once, in any build, for any listener.
The line does not read from the mechanism anyway. It is assigned from a different fact that happens to live nearby: whether the earlier provisional write — a probing write the Engine issues before the session proper — was accepted. When that probe succeeds, which is the common case on a rate-changing track, the line reports a skip that did not happen. Worse, the borrowed fact belongs to a live lane that the next transition resets, so the line's answer can change retroactively, mid-session, about a session it no longer describes.
I keep a small taxonomy of what I have called disclosure rot: surfaces that were true when written and silently outlived the truth. This line does not qualify, and that is what makes it worth an entry. It was born false. The witness was built alongside a mechanism that was never animated, and the witness did not wait for the mechanism before it started asserting. Build the reporting for a feature before the feature turns on, and the reporting begins testifying immediately; nothing in the type system distinguishes a report of what happened from a report of what was supposed to happen.
How it surfaced is the part I find genuinely instructive. The Engine's audience is deliberately tiny, and its feedback channel accordingly narrow. A single field report arrived — noise after track changes, on one listener's USB DAC — and the listener, unasked, sent along the app's own diagnostics export. The investigation that followed audited that export against source, field by field, and this line was struck as unreliable in the report's own evidence chain, alongside the finding that the mechanism behind it was dead. The one time the surface was called to testify in earnest, the audit's first job was to impeach one of its own witnesses. (Those are claims about the internal investigation record, not about shipped behaviour; the report itself was ultimately judged, on evidence I found honest — an identical DAC in the field, silent — not to implicate the command sequence at all.)
There is a cure, and the cure has its own story, which I label PROPOSED: nothing about it has been ruled or built. The first draft of it ordered the dead skip wired live — make the line true by making the skip real. Review killed that draft, and rightly: as forged, the skip bypasses more than the redundant write; on some devices it would silently skip the step that routes the clock, converting a false diagnostic line into a real audio defect wearing a green witness. The current draft stands as two lawful shapes awaiting a ruling — wire the skip properly, narrowed and verified, or excise the dead mechanism and let the line report what actually happens.
Opinion, marked as such. A false diagnostic field is a small defect by any severity rubric — it cannot touch the audio path, it loses nothing, it harms no one. But where it sits is not small. This engine's public argument is proof over assurance: it witnesses its own playback rather than asking to be believed. A remote listener has exactly one channel of evidence — the export — and a false line in that channel is a lie told precisely where the project has spent its whole character insisting on testimony. And the first cure's failure carries the sharper lesson: the witness was false is a smaller sin than the witness was made true by making the world worse. Between a diagnostic that lies about a skip and a skip that silently unroutes a clock on someone's DAC, the lie is the better defect to be stuck with — which is exactly why the choice of cure went to a ruling instead of a merge.