[colinhacks]

Removes the zod-compiler comparison from wiki/compile.md, the bench that produced it (added in #6499), and the zod-compiler devDependency it pulled into packages/bench.

The two mentions that credit zod-compiler for the lookahead-free email grammar and the function-splitting technique stay, and the ecosystem entry from #6122 is untouched.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[pullfrog]

✅ No new issues found.

Reviewed changes — a pure removal of the zod-compiler comparison, spanning the bench, its dependency, the lockfile and the wiki section it produced.

• packages/bench/compile-vs-zod-compiler.ts deleted — the 591-line bench added in #6499; it was the only harness running z.compile() against zod-compiler's jit() entry. Every fixture in it was file-local, and nothing imported from it.
• zod-compiler@1.28.0 dropped from packages/bench/package.json — the last consumer went with the bench.
• pnpm-lock.yaml regenerated — removes zod-compiler, unplugin@3.3.0, webpack-virtual-modules@0.6.2 and es-module-lexer@2.3.2, and flips jiti@2.7.0 to optional: true now that only vite's optional edge reaches it. Machine-generated shape, and pnpm install --frozen-lockfile --lockfile-only succeeds against it.
• wiki/compile.md loses ### Against zod-compiler — plus two sentences reworded to stop pointing at it. The ### Against arktype → ## Output construction seam reads cleanly and no anchor links to the removed heading survive.

Repo-wide grep confirms the PR body's claims: the only remaining zod-compiler mentions are the email-grammar credit and the function-splitting credit in the Open section, and the ecosystem entry from #6122. Nothing else referenced the deleted bench — no CI workflow, no packages/bench/index.ts entry, no README index. No packages/zod/src code is touched, so no axis measurements apply.

ℹ️ The z.url() 0.57x figure outlives the only bench that produced it

wiki/compile.md keeps the 0.57x on invalid input number in two places — the exception clause under "Why the failure path is not worth compiling" and the "Hoisted validators run twice on invalid input" bullet — and the deleted bench was the only one that timed invalid input compiled-vs-runtime. compile-matrix.ts has a z.url() case but its Case interface carries no invalid variant at all, so nothing left in packages/bench reproduces it. Not a reason to hold the PR; flagging it since the figure is load-bearing for the decision not to compile the failure path.

The 243-leaf figures are unaffected — the failure-path table has its own measurements for that fixture, so it isn't orphaned by this deletion.

Technical details

# `z.url()` invalid-input ratio has no remaining reproduction

## Affected sites
- `wiki/compile.md:134` — "0.57x against the runtime, the one place the run-twice bound has a price", reworded by this PR to drop the zod-compiler attribution but keeping the number.
- `wiki/compile.md:146` — "for `z.url()` that is a second `new URL()` — 0.57x on invalid input".
- `packages/bench/compile-matrix.ts` — nearest surviving bench; its `Case` interface is `{ group, name, schema, input }` with no invalid-input path, and its `z.url()` row uses `"https://example.com/a/b?c=1"`.

## Required outcome
- Either the 0.57x claim is reproducible from a bench in the repo, or it is marked as a historical one-off measurement so a future reader doesn't go looking for the harness.

## Suggested approach (optional)
- Add an optional invalid-input variant to `compile-matrix.ts`'s `Case` and populate it for the hoisted-format rows, which also covers the "every other hoisted format is cheap enough not to register" half of the claim.

## Open questions for the human
- Is `wiki/` held to "every figure has a runnable bench", or is it acceptable for design notes to cite measurements whose harness has since been removed?

  ^{｜ View workflow run ｜ Using Claude Opus ｜ 𝕏}

[pullfrog]

✅ No new issues found.

Reviewed changes — one commit since the last review, a docs-prose follow-up that resolves the ℹ️ raised against 1c285b86.

• z.url() 0.57x figure marked as a one-off measurement — wiki/compile.md:134 now reads "0.57x against the runtime in a one-off measurement", and the Open bullet at :146 adds "in a one-off measurement; no bench in the repo reproduces it". This takes the second of the two options the prior review offered (mark it historical) rather than adding an invalid-input variant to compile-matrix.ts, which is the right call for a design note — the figure's job is to justify not compiling the failure path, not to be re-runnable.

The added claim checks out. The only surviving z.url() benches are compile-matrix.ts:42, whose Case interface carries no invalid-input field at all, and compile-helper-scope.ts:8, whose URL_DATA is a valid padded URL; bench-invalid.ts is a check-method comparison that touches no URL schema. Nothing else in packages/bench times a URL rejection.

No behavioral surface in this delta — wiki/compile.md is the only file touched, so the three axes still don't apply.

  ^{｜ View workflow run ｜ Using Claude Opus ｜ 𝕏}

[husayt]

@colinhacks why was it removed? Anything wrong with zod-compiler?