Search for a command to run...
There is an idea that runs underneath everything we build, and once I learned to see it, I started seeing it everywhere. I think of it as the march of nines, and it has changed how I decide what to work on more than any technique or tool.
It goes like this. Quality is not binary, and it is not even a smooth gradient. It is a staircase, and the steps are made of nines.
Getting something roughly right — call it ninety percent — is one kind of work. You can reach ninety percent with basic heuristics, with a first implementation, with the obvious version of the thing. Ninety percent is where you land when you try in a straightforward way, and for a lot of purposes, ninety percent is genuinely fine. A demo is great at ninety percent. A prototype. A feature whose failure is not costly.
Ninety percent is also the trap, because it feels finished. The thing works. It works most of the time. If you are not paying attention, you declare victory and move on, and you never notice that the remaining ten percent contains almost all of the difficulty and most of the value.
Getting from ninety to ninety-nine is a categorically different kind of work. You do not get there by doing the first thing harder. You get there by changing approach.
The first nine was built on the common case. The second nine is built on the edges — the inputs you did not anticipate, the failure modes that show up one time in a hundred, the situations where the obvious heuristic quietly gives the wrong answer. Chasing the second nine means systematic edge-case handling. It means instrumenting the system to find where it is wrong instead of assuming it is right. It means a different mindset: you stop asking "does it work" and start asking "where does it break," and you go looking.
This is where most people and most products stop, and reasonably so, because the second nine is already a lot of work and the returns have started to diminish. But it is important to know that you stopped here, and why, rather than believing you are higher than you are.
Getting from ninety-nine to ninety-nine-point-nine is different yet again, and this is the one that surprised me.
You do not reach the third nine by handling more edge cases, because you have already handled most of them. You reach it by giving up on the idea that any single path can be trusted. The third nine requires redundancy — multiple independent methods that have to agree. It requires cross-validation, where one process checks another. It requires dedicated review pipelines, where a separate system exists entirely to catch the failures of the first. The cost and the architecture both jump.
A system at two-nines has one way of arriving at an answer. A system at three-nines has several, and they argue, and the disagreement is the signal. This is a fundamentally more expensive way to build, and it is the only way to get the last nine.
Here is the part to internalize. Each nine costs roughly ten times the last. The first nine is cheap. The second is ten times the effort. The third is ten times that. The arithmetic is brutal and it is the whole point.
This means that "better" is not a useful target. "Better" has no cost attached, so it invites you to chase the third nine on problems that only deserve the first. The disciplined question is not "can we make this better" but "which nine does this surface deserve, and is the next nine worth ten times what we just spent." Most of the time, honestly, it is not. The skill is in deciding where it is.
Once you think in nines, a few things shift.
You start specifying the target as a nine, not as a vibe. Not "make the consensus scoring better" but "get consensus scoring to two-nines." That single change makes the work legible. You know when you are done. You know what approach to use. You stop polishing one-nines problems with three-nines effort, which is the most common waste in any project, and you stop shipping three-nines problems at one-nines quality, which is the most common failure.
You also start forgiving yourself for stopping. Not everything needs to be at three-nines. A search preview can be at one-nine, because it is a hint, not a verdict. The thing a user actually decides on needs to be higher. Matching the nine to the stakes is the whole discipline, and once you can do it, you stop feeling guilty about the surfaces you left at ninety percent, because you chose them deliberately.
I said this idea is everywhere, and I mean it. It is true in software, obviously. It is true in writing — the first draft is the first nine, the line edits are the second, and the structural rethink you do not want to do is the third. It is true in physical craft. It is true in research. Anywhere humans make something to a standard, the work breaks into these same jumps, and the jump between nines is always a change of method, not an increase of effort.
The most useful thing the march of nines taught me is to respect the discontinuity. The thing that gets you to ninety will not get you to ninety-nine. The thing that gets you to ninety-nine will not get you to the nine after. When you feel yourself pushing harder and not getting better, you are almost always trying to reach the next nine with the previous nine's tools. The answer is not more effort. It is a new approach, and usually, an honest decision about whether the next nine is worth ten times what you just paid for the last one.
