Execution vs strategy in algo trading — what actually fails more often?

I’ve been trying to break down why so many trading systems fail in real usage, even when the underlying strategy looks solid on paper.

From what I’ve seen, the issue is often not the model or signal logic itself, but the execution layer:

risk parameters changing mid-way
inconsistent position sizing
manual intervention during drawdowns
“strategy drift” over time
emotional overrides of automated rules

It feels like a lot of edge gets lost not in generation of signals, but in how consistently those signals are actually followed in production.

In fully systematic setups, do you find that most failures come from:

strategy design flaws
execution/risk layer issues
infrastructure / latency / data problems
regime changes

Curious how others here split this in practice.