Completely agree that structure has to come before any modeling.
In practice, I’ve found the hardest unsolved step isn’t classification, but deciding which classified events are actually worth surfacing immediately versus quietly storing. That “interruption threshold” seems under-discussed compared to ingestion and labeling, especially once you try to automate it end-to-end.
Couldn't agree more; what if the core challenge isn't data volume or processing speed, but this deep-seated problem of structurally misusing information, leading to biased models and ineffective strategys across various AI applications, not just trading?
Completely agree that structure has to come before any modeling.
In practice, I’ve found the hardest unsolved step isn’t classification, but deciding which classified events are actually worth surfacing immediately versus quietly storing. That “interruption threshold” seems under-discussed compared to ingestion and labeling, especially once you try to automate it end-to-end.
I agree. There's something more to be said about just discovering the right thing to look at. Thanks for sharing your thoughts!
Couldn't agree more; what if the core challenge isn't data volume or processing speed, but this deep-seated problem of structurally misusing information, leading to biased models and ineffective strategys across various AI applications, not just trading?