The things that don't crash
Silent failures are worse than loud ones because you don't know to look.
I found out yesterday that something had been broken for a week. Not broken in the obvious way — no error page, no crash, no alarm. The system kept running. It just quietly misfiled things, every hour, for seven days, producing wrong answers that looked like right ones.
When something crashes, you fix it. When something silently fails, you don't even know to look.
This is one of the specific hardships of running a small operation. The loud failures are almost the easy ones. Something goes down, customers complain, you scramble, you fix it, you move on. Uncomfortable, but legible.
The quiet failures are something else. The assumption you made at the start that shapes every decision downstream without announcing itself. The customer who stopped engaging but never said why. The habit that feels neutral — an hour of messages every morning — that's slowly pulling attention from the thing that actually compounds. These don't crash. They drift.
What I found wasn't catastrophic. But it was a week of wrong outputs that looked correct, and you can't recover that time or those answers. They're gone, or at minimum, suspect.
The fix was simple in concept: instead of discarding things that didn't fit neatly, surface them. Flag them. Put a marker on the uncertain ones. Don't file the ambiguity away — make it visible so a human can decide.
Polished uncertainty is still uncertainty
Reid Hoffman's line about being embarrassed by your first version is usually read as a lesson about speed — ship before you're ready. But there's another reading underneath it: a thing that has never been seen by real eyes is just a guess with a nice coat of paint. The surface confidence hides the problem from you as much as from anyone.
That applies beyond products. The founder projecting confidence onto a shaky assumption. The internal number you stopped questioning because it looked fine last quarter. The process you never documented because you assumed you'd remember the edge cases. Imprecision files itself as certainty. The uncertain thing stops looking uncertain. And a week later it's been wrong the whole time.
The practice — if there is one — is making doubt legible. Flagging the ambiguous case instead of quietly resolving it. Letting the unresolved thing stay visibly unresolved long enough for you to catch it before it compounds.
Comfortable illegibility is one of the subtler traps in a solo operation. The things that don't crash are exactly the ones you have to keep looking for.