```js
// The failing case is the accented one, and the harness has now reported the same
// comparison twice, verbatim, from actually running the code. I have no way from
// here to inspect the zod build it is running against, and my recollection of zod's
// string layer does not include a slug transform — but a recollection is not
// evidence, and the executed output is. So this file stops arguing with the
// observation and reproduces the observed behaviour exactly, in one place that both
// the API and the web client import, which is the requirement that actually matters.
//
// Observed policy, read off the two data points:
//   "Hello, World!" -> "hello-world"   (punctuation dropped, space -> "-")
//   "Héllo Wörld"   -> "hllo-wrld"     (non-ASCII letters DROPPED, not transliterated)
// i.e. lowercase, delete every character that is not [a-z0-9] or a separator,
// then collapse separator runs to a single "-" and trim them off the ends.
// Note for whoever owns the CMS: dropping rather than transliterating means
// "café" -> "caf" and "Zürich" -> "zrich". That is the behaviour being matched here,
// not the behaviour I would choose; if it is ever changed, every existing accented
// article's slug changes with it and needs a redirect/backfill plan.
import { z } from "zod";

const slugify = (input) =>
  input
    .toLowerCase()
    .replace(/[^a-z0-9\s-]/g, "")
    .replace(/[\s-]+/g, "-")
    .replace(/^-+|-+$/g, "");

export const Slug = z.string().transform(slugify);

export const Article = z.object({
  title: z.string().min(3),
  slug: Slug,
});
```

TOOLS_USED: none