If they are going to build on a beach, clearly they should be using the Seabees. I joke but the situation is still awful.

I would love to hear strategies for dealing with this. I have been on the no one side before and sometimes it is like shouting into the void.

Having worked alongside them, I can guarantee there is a story and a name for this. I don't happen to know it though.

But when looking at how my clock is performing vs the monitors worldwide, I get some interesting results. Practically everything is within +/- 5 ms. However, there are these spikes in offset every day only for 3 European monitors in NL and DE at 0100-0300 UTC. The others are fine.

I've absolutely done this, but I also measure a host of metrics in parallel with a set of prompts. Latency, cost, and reliability are also worth considering. For some applications multiple models are used based upon performance.

If you want to do some rigorous evaluations over a set of inputs, MLFlow worked well for me with the different services over API. I wrote a bunch of metrics for latency and cost for the services and was able to argue which model was the best value for our use case with that data.

Shower thoughts: if you survive an encounter you will upgrade your equipment :D Also, the gear is a threat to those who would challenge you. We are starting with fast fashion and ending up with the clothes which last (which might be the tacticool leather).