The only number we could see was the one that lied to us.

Donella Meadows taught me a distinction I now can't stop seeing: stocks versus flows. A flow is the rate of something moving, payments per minute, features shipped this sprint. A stock is what's quietly accumulated underneath, trust, reliability, debt and the one I learned the hard way: knowledge. Our minds fixate on flows, because flows are visible and loud. Stocks change slowly and silently, so we ignore them until they force themselves into view.

Pre-Covid, our payment processing engine was powered by a codebase that was slowly losing touch with the rest of us. A small team had built it and built it well, on Akka.NET and the actor model, properly deep in distributed-systems thinking. The concepts were genuinely brilliant. The problem wasn't the code. The problem was that the understanding of it lived in a few heads and never became institutional. It was a stock sitting in people, not in the organization.

Then the people who understood it moved on. And the stock walked out the door with them.

What we were left with was a system almost nobody remaining could fully read. We hadn't been tracking the thing that mattered, our collective grasp of how the engine actually worked, because it never showed up on any dashboard. The only thing we could measure was the flow at the very end: successful versus failed payments in the database. That number was all we had and it told us almost nothing. We knew to restart the service when it timed out on an external call, or crashed, or just ate memory until it fell over. We were flying by one dial, with no idea what was happening inside the engine that drove it.

That experience changed how I think about visibility and it pushed two decisions. We moved to microservices, yes. But the more important one, the one I'd defend harder, was making instrumentation a standard in every codebase. StatsD and Grafana then, matured into Datadog now. For the first time we could actually see the system, input to engine to output, not just the score at the end. We stopped guessing from a single number and started listening to what the system was telling us about itself.

Meadows' point is that a system you can only see the output of is a system you can't manage, you can only react to it. A green success rate felt like health. It was actually the one number still working while everything we'd failed to make visible drained away behind it. The dangerous stocks are always the invisible ones. The knowledge that isn't written down. The reliability you're spending without replacing. The trust quietly leaking while this week's numbers still look fine.

So now I ask a different question. Not just "what's moving," but "what's accumulating, or draining, that nobody's looking at?" Because the flow will tell you everything's fine right up until the moment it very much isn't.