Someone read ten pages of a story my writing system had produced and came back with four complaints. The place was hard to picture. She couldn't tell who a character was talking to. One phrase, a specific number of years, showed up about ten times. And the sentences didn't flow.
The automated review had read the same ten pages and passed them. All good.
The interesting part isn't that the review missed something. It's what the review was built to miss, and why I built it that way.
A fence around an empty field
I went back and read my own style rules like a stranger.
Maximum twenty-five words per sentence. Never chain more than two clauses. No weather or setting descriptions unless they affect what happens. Skip physical descriptions of things the character already knows well. No metaphor or simile by default. Sensory inventories banned.
Every one of those is defensible. Every one was written in response to something real, a specific paragraph that made me wince, a habit I'd watched go wrong.
Read as a set, though, they say something I never meant to say. They ban cadence. They ban place. They ban figurative language. What's left is short declaratives about actions taking place in an unspecified room.
I had written down forty things prose must not do and almost nothing about what good prose does.
Prohibitions are enforceable, which is why they win
Here's the mechanism, and I think it generalizes well past writing.
"Never chain more than two independent clauses" is countable. You can check it in one pass. You can hand it to a collaborator, an editor, or a machine and get consistent compliance without any judgment involved.
"Make this paragraph flow" is not countable. It requires taste, and taste doesn't survive being written on a checklist.
So the checkable rules accumulate and the uncheckable ones quietly never get written down. Not because you value them less. Because there's nothing to write. A few years of hard-won lessons later, your style guide is all fence and no field.
There's a second reason bans win: you only write rules for failures you notice. Purple prose announces itself. Flatness doesn't. Nobody reads a page of clean short sentences and thinks something is wrong here. They just stop reading, and they don't file a complaint about it.
Starve a writer of tools and watch what's left
The repetition was the part that actually changed how I think about this.
That phrase came back ten times, and I'd been treating it as a discipline failure. It wasn't. Emphasis has to come from somewhere. If you want a fact to land with weight, you can build a long sentence that accumulates toward it, or reach for a metaphor, or elaborate. All three of those were banned.
The only remaining way to make something feel important was to say it again. So the system said it again, and again, and again. Repetition wasn't sloppiness. It was the last legal tool in the box.
My banned-constructions list even sanctioned it. One-line portentous paragraph, cap two. A cap of two is not a prohibition. It's permission, in writing, for the exact crutch I was annoyed about.
Rewriting bans as intent
The fix wasn't deleting rules. It was rewriting them so they name the target, not just the ditch.
The sentence cap became a rhythm rule. A run-on chains three or more clauses to sound literary, so cut those, but a paragraph of "He did X. He did Y. He did Z." is mechanical. Let a longer sentence carry a continuous action or thought, then let a short one land the emphasis.
The setting ban became a grounding rule. The reader has to be able to picture where they are. Give the geography, the scale, one or two textures that make it real. A character who knows a place well still moves through it, so render it through movement instead of a static catalog.
The sensory ban got narrowed to what I actually objected to. Ban the inventory, not the sensing. Two chosen details are good. A checklist of smell, light, and air temperature is not.
Same underlying judgment in all three cases. The difference is that the new version tells you what to aim at and where the ditch is, instead of only where the ditch is.
The review has to change too
Rules only matter if something checks them, and my checker had the same disease.
It was asking two questions. Can a stranger follow the plot? Are the invented nouns explained before they're used? Both answers were yes, so it shipped. Comprehension was the only dimension it had, and flat prose is perfectly comprehensible.
It now asks four more. Can you picture the space. Do you know who's speaking to whom. Does the rhythm vary. Is anything repeated past its usefulness. Those aren't comprehension questions, they're reader-experience questions, and they're the ones an actual reader asked first.
There was one more thing worth flagging. The pass that applies notes had been instructed to use the lightest possible touch and not to add warmth. Which meant that even when a problem got flagged, fixing it was out of bounds. I had built a system that could diagnose without permission to treat.
The takeaway
Go read your own rules the way a stranger would. Your style sheet, your team's review conventions, your editing checklist, the notes you leave yourself at the end of a session.
Count how many of them begin with no, never, avoid, or don't. If that's most of them, you haven't written down your standards. You've written down your fears. The list describes the maker you're afraid of being rather than the one you're trying to become.
Both are worth writing down. Only one of them usually makes it onto the page.