I joined and for kicks entered most of my audio gear and asked for an analysis. Your AI needs tuning. It screwed up the specs of about a third of the items (admittedly, many are vintage), so I went back in the notes with annotations. It still screwed up a couple. And, apparently, my weakest link is Michael Fidler's MM Pro Mk II - probably one of the finest phono stages in the world. But, it's kind of fun watching it clutch at straws. Also, it didn't like my Sony UBP X800 Mk2 that I use only as a CD & SACD transport (in my notes). It suggested a $5000 transport that does not play SACDs. Hey, it's AI; maybe it will learn.
Boy, is it! I tweaked my notes in an attempt to correct some of its misunderstandings an it got worse: it wants me to drive my monoliths IIIs with their non-existent internal crossover using the Apt model 1 power amp I have connected to my motion 15i speakers. Part of the problem is that I have what is essentially two complete systems but they are interconnected. It can't compute. But it is amusing.
I have updated the AI instructions and this explains what was happening and how it will work going forward. I appreciate your great feedback and if you can stick with me, we might achieve something better. I will probably release an update later today or this evening.
Re: System Analysis — What's Happening and What's Coming
A few things to explain about what you experienced:
On incorrect specs, especially for vintage gear
The AI powering System Analysis (Google's Gemini) has a knowledge cutoff of early 2025, and its training data on vintage and niche components can be thin or flat-out wrong. It knows a Schiit Bifrost better than it knows a 1978 Marantz. The Notes field exists precisely so you can correct it — and your instinct to annotate was right. The next build tightens this significantly: the AI will be explicitly instructed to treat your notes as ground truth and not contradict them with its own assumptions.
On the "weakest link" misfire
If the AI doesn't recognise a component — a bespoke phono stage, a rare vintage piece — it has been defaulting to flagging it as a weakness simply because it doesn't know it. That's backwards. The next build adds a rule: unfamiliarity is not a weakness. The AI will only identify a weak link when it has genuine knowledge of the component, and will acknowledge when it can't make a confident call.
On the two-system topology confusion
This one is a structural limitation of the current format. A flat list of components doesn't tell the AI what's connected to what, so it guesses — sometimes badly, as you found. The next build instructs the AI to map out signal chains from your components and notes before writing anything, and to admit ambiguity rather than invent connections. In the meantime, explicit wiring notes on each component help a lot — something like
"connected to [speaker name] only" gives it enough to work with.
What's already working in your notes
Your notes are being sent to the AI right now. The next build makes them authoritative. So the more explicitly you describe your setup in notes, the better the analysis will be even today — and significantly better with the next update.
Your setup — two interconnected systems with carefully curated vintage and specialist gear — is genuinely at the challenging end of what any AI analysis tool can handle. Appreciate you sticking with it.
