# AI writing filter — accumulated knowledge

Append-only across sessions. Each entry records a specific stylistic
tell that showed up and how it was rewritten.

## Format

```
## YYYY-MM-DD — <short title>
Story: <story-id>
Pass: <pass number>

<the pattern, before/after example, and the rule>
```

---

## 2026-05-30 — "Both X are real X. They answer different questions."
Story: punching-above-50
Pass: 9

The biggest single tell. The two-step is: a parallel "A finds X, B
finds Y" construction followed by the earnest closer "Both are valid"
or "Both lists are real lists" or "They answer different questions."
Rewrite by collapsing into a single sentence and dropping the
emphatic closer.

**Before:**
> A ranking that adjusts for student backgrounds finds the schools
> doing the most work with the kids they have. A ranking that doesn't
> adjust finds the schools admitting the kids who'd score well
> anywhere. Both lists are real lists. They answer different questions.

**After:**
> The two kinds of ranking are answering different questions — one
> asks which schools get the most out of their students, the other
> asks which schools admit the highest-scoring students. A parent
> comparing options usually wants to see both.

## 2026-05-30 — Smuggled value judgment about who matters
Story: punching-above-50
Pass: 9

The pattern: writing that treats catching-up-to-proficient as the
only valuable kind of growth, and dismisses wealthy-area schools as
having "no room to grow." False on its face — proficient is a low
bar, and strong students can keep growing. When the prose implies
that one kind of student's growth matters more than another's,
rewrite to describe the data without ranking the kids.

**Before:**
> Wealthy-area schools rarely show up on this list — they don't need
> to score above expectations because their expected score is already
> at the ceiling.

**After:**
> Wealthy-area schools rarely appear on this particular list because
> their proficiency rates are already near the top of the scale, so
> the gap-above-expectation measure doesn't pick them up well. A
> different ranking — gains on advanced measures like Regents pass
> rates, AP performance, or top-scoring share — would surface a
> different set of schools, including many wealthy-area ones still
> helping strong students grow.

## 2026-05-30 — "Confirmed." as a sentence-starter
Story: multiple
Pass: 9

The opening word "Confirmed." (followed by a period and the actual
finding) is a strong model tell. The reader doesn't need to be told
the analysis confirmed the thesis; they're about to read the
analysis. Cut "Confirmed." and start with the finding.

## 2026-05-30 — "It is essentially uninformative" / "essentially zero"
Story: multiple
Pass: 9

"Essentially" plus a sweeping characterization is a model favorite.
It usually overstates. Replace with the actual measurement: "the
correlation is 0.11" (or "close to zero") rather than "essentially
uninformative." If the claim is genuinely sweeping, state the
number and let the reader see how sweeping it is.