Go to the source, or don't report at all
Most bad decisions aren't from bad judgment. They come from trusting a summary when the source was one step further away.
There was an automated alert that told me nobody had shown up to a class. The alert was wrong. People had shown up. The tool was reading from a manual transcription layer instead of going to where the actual record lived.
It sounds small. A quick explanation, a few confused messages. But the cost of a false alarm isn't the confusion. It's trust. Once a system cries wolf, everyone who received it starts filing future alerts under "maybe." The next true one lands softer.
The root cause was one layer of indirection. Someone had copied data into a secondary location. The system trusted that copy. The copy was stale.
The fix: read the original source directly. And add one more rule - don't report unless you can verify you're reading what you think you're reading. If the check fails, stay quiet.
That last part is the one worth sitting with.
Most bad decisions I've made as a founder weren't from bad judgment. They were from trusting a summary. A secondhand market take. A dashboard that was counting the wrong thing. A customer quote I remembered differently than it was said. A number someone told me was "roughly accurate."
The source of truth was always one step further than where I stopped.
Deliberate silence is underrated. It feels like paralysis, but it's closer to discipline. The alternative is filling the gap with a guess, calling it a decision, and occasionally being right by accident - which is worse than being wrong, because it trains you to trust the gap.
Pieter Levels says you don't need a co-founder, you need a customer. I read this as: stop seeking confirmation from sources that already agree with you. The customer is the source. The market is the source. Go there, not to the interpretation of it.
The same logic runs inside a business. The source of what's actually happening is never the report. It's the raw record. The conversation. The original page, loaded fresh, read directly, verified before acted on.
When I build automation now, I try to give it one governing rule: can you prove you're reading what you think you're reading? If not, wait. Don't guess. Don't fill the gap with a conclusion that sounds plausible.
I'm trying to learn to ask myself the same question before I decide anything.