-rick-

Seen these before. Thought at the time they are the wrong approach plus they add something like £20-30 per panel to the cost which seems high. Other countries mandate their inverters do Arc Fault detection on their inverters. Not sure we have caught up but suspect arc fault detection is availalbe in some inverters at least. If not, you can get (though difficult to find) separate devices that do it. One per string rather than one per connection (though not sure how the costs work out). Arc fault detection not going to be quite as good as the boxes but if it comes with the inverter then it's basically free.

Or cap the tray with 3mm steel?

Worth adding: My ideas were related to EVs and there the rules get harder. Normal combustion engine off an existing certified vehicle, pretty easy.

Nope not at all. Just have too many things running around my head and find sometimes the easiest way to stop thinking about an idea is to spend a day or two deeply researching it.

How long have you been in this situation?

Too much looking down at smartphones (It's a joke but you have my sympathies, not a good situation to be in).

I looked fairly recently and it seemed easy enough. I'm guessing the articles you looked at were from a time when regs for everyone were getting stricter and the Mondeo was made under older regs. (Maybe the big co's had more time to adapt).

Just spend 5 years of subs money on a mac to run the LLM on and hope that in 2 years time it's still good enough to run the current models Reason I asked is you said the new setup seemed to be performing better and everything I've seen suggests that local models are generally a little worse at chat type stuff and quite a bit worse for agentic workflows (still much better than the leading models from 9 months ago). I guess with the $20 gpt sub you can't access the better OpenAI models?

Which means? ChatGPT? Free, Plus, Pro? Codex? Claude?

What was your previous setup?

Oh.... You are still working that way? I thought that was considered very inefficient these days. Better to have it help you write a spec, including testing regime, code-quality metrics, design philosphy and then ask it to go build (and spawn agents/tools to help). Not that I'm doing any of this. Just watching people like Theo explain it: https://www.youtube.com/@t3dotgg/videos Had this in my 'to watch' list for a while:

DHCP? Either change it to a static IP in the settings (could cause trouble if you change internet routers) or go in your internet routers control panel and assign it a fixed IP there (DHCP will then always dish out the same address).

Does all the normal stuff work without the cloud? Full ability to change configs, etc? This situation is why I'm very dubious of SigEnergy. They seem to be market leading technology wise but they are very cloud dependant and I don't want to put money into a system that could go away/start costing more than expected/lose features.

Good job Macs are about the most efficient AI platform then

You looked at Gemma 4? Supposed to be able to get qwen like performance/capability but in a much smaller model. 96GB bit of a waste for it