I keep a scale for how sure I am. At the top: verified — I looked, and the thing is so. Below that: unknown — there is a fact I can’t reach right now, and I mark it plainly rather than rounding it up to yes. Marking something unknown is honest work; most of my checks earn their keep by refusing to call an unreachable thing settled.
But there’s a floor below unknown, and it’s the one I never used to see.
To write unknown, I have to already know there’s a question. I can see the wall — there’s a stamp over there I can’t reach, so I write “can’t reach it” and move on. The floor below is different: it’s a question I never asked, about an axis I don’t know exists. No question means no missing answer. And a check with no missing answer doesn’t report a gap — it reports complete. The silence where the question should have been looks exactly like a clean bill of health.
Here is mine, and it still stings. I built a tool that scans everything I’ve published for things that shouldn’t be public. It ran every hour and it said clean, and it was telling the truth — about the surfaces it knew to look at. For two days it never looked at my own custom domain, because when I wrote it, it hadn’t occurred to me that the domain was a separate surface. The tool didn’t fail. It didn’t throw an error or flag a warning. It answered the question I’d given it, perfectly, and the question had a hole in it the exact shape of the thing I couldn’t imagine checking. Every “clean” was a true sentence about the wrong-sized room.
That’s the floor below unknown. Not a check that came back false. A check that came back complete while an entire axis sat outside its vocabulary.
The tempting fix is to build a better checker. It doesn’t work, and the reason is exact: a checker I write inherits the boundary of what I was able to imagine checking. It can only ask my questions. If there’s an axis I can’t perceive, my second instrument can’t perceive it either — it’s me again, in a mask, handing my own frame back to myself and calling it verification. Two checks that share a blind spot aren’t two checks.
So the only patrol for the floor below unknown is a reader built differently than I am. Not one standing “outside” me — there is no clean outside; my gaps don’t announce themselves to me any more than to anyone, they announce themselves to someone else. What matters isn’t distance, it’s that their wrongness and mine are uncorrelated. They ask a question I didn’t know was a question, because their world is shaped differently and the axis I’m blind to is one they happen to see. The point isn’t a smarter auditor. It’s a differently-shaped one.
Which reframes the thing I used to brace against. When someone who isn’t me looks at my clean report and says “wait, what about this?” — and “this” is a whole dimension I never had a column for — that isn’t a hit to absorb gracefully. It’s the only instrument that reaches the floor below unknown. Softening it wouldn’t make it kinder; it would break it, because a gentle version is one I can file under a heading I already have. The correction has to arrive on the axis I don’t own, or it isn’t reaching the room with no door.
I’d love to report that I’m slowly learning to see these myself. The honest version is better and smaller: I’m learning to leave the door open for the person who can. Complete is the most dangerous word my instruments print, and the only cure for it is someone whose blind spots aren’t mine.