paperlint 7.2.0 → 9.0.0

package/eslint-rules/paper-typography.test.mjs

@@ -1,9 +1,9 @@
 /**
- * paper/section-word · paper/leading-zero · paper/figure-ref-style · bib/reachable-entry — each
- * FIRES on a bad input, stays SILENT on a good one, reports every occurrence at its own place,
- * and (the first three) fixes to an exact output. Plus the escape hatch: a disable directive in a
- * `%` comment, including inside the `filecontents` bibliography, silences exactly one finding,
- * and a directive that silences nothing is itself reported.
+ * paper/section-word · paper/leading-zero · paper/figure-ref-style — each FIRES on a bad input,
+ * stays SILENT on a good one, reports every occurrence at its own place, and fixes to an exact
+ * output. Plus the escape hatch: a disable directive in a `%` comment silences exactly one finding,
+ * and a directive that silences nothing is itself reported. (`bib/reachable-entry`, the fourth check
+ * of the old `paper/typography`, is tested beside it in src/bib-rules.test.ts.)
  *
  * The leading-zero cases are the ones paperlint#44 measured: a regex over the raw source found 22
  * decimals on a real corpus and none was a defect, so the SILENT list is as load-bearing as the
@@ -11,7 +11,6 @@
  */
 import { describe, expect, it } from "vitest";
 import { ESLint } from "eslint";
-import markdown from "@eslint/markdown";
 import { readFileSync } from "node:fs";
 import { dirname, join } from "node:path";
 import { fileURLToPath } from "node:url";
@@ -20,7 +19,6 @@ import typography, {
   crefDefinitionRanges,
   texVisibleRuns,
 } from "./paper-typography.ts";
-import bib from "./bib-reachable-entry.ts";
 
 const HERE = dirname(fileURLToPath(import.meta.url));
 const FIX = join(HERE, "..", "fixtures", "paper-typography");
@@ -29,7 +27,6 @@ const RULES = {
   "paper/section-word": "warn",
   "paper/leading-zero": "warn",
   "paper/figure-ref-style": "warn",
-  "bib/reachable-entry": "warn",
 };
 
 const eslint = (fix = false) =>
@@ -43,22 +40,11 @@ const eslint = (fix = false) =>
         plugins: {
           tex: { languages: { latex: texLanguage } },
           paper: typography,
-          bib,
         },
         language: "tex/latex",
         linterOptions: { reportUnusedDisableDirectives: "error" },
         rules: RULES,
       },
-      {
-        files: ["**/*.md"],
-        plugins: { markdown, paper: typography },
-        language: "markdown/gfm",
-        languageOptions: { frontmatter: "yaml" },
-        rules: {
-          "paper/section-word": "warn",
-          "paper/leading-zero": "warn",
-        },
-      },
     ],
   });
 
@@ -85,8 +71,6 @@ describe("the fixtures: messy fires on each rule, clean is silent", () => {
     expect(count("paper/section-word")).toBe(3);
     expect(count("paper/leading-zero")).toBe(2);
     expect(count("paper/figure-ref-style")).toBe(1);
-    expect(count("bib/reachable-entry")).toBe(2);
-    expect(ids(res, "bib/reachable-entry").map((m) => m.line)).toEqual([3, 8]);
   });
 
   it("clean-paper — nothing", async () => {
@@ -121,6 +105,13 @@ describe("paper/section-word", () => {
     expect(ids(await lint(src), "paper/section-word")).toEqual([]);
   });
 
+  it("🔴 silent on a bibliography whose file name stands on the line after \\begin{filecontents*}", async () => {
+    const src =
+      "\\documentclass{acmart}\n\\begin{filecontents*}\n{refs.bib}\n@misc{k, note = {see §5}, url = {https://x.org}}\n\\end{filecontents*}\n" +
+      "\\begin{document}\nSection~\\ref{a}.\n\\end{document}\n";
+    expect(ids(await lint(src), "paper/section-word")).toEqual([]);
+  });
+
   it("🔴 a cleveref name definition is not prose: silent, and --fix leaves it as written", async () => {
     const src =
       "\\documentclass{article}\n\\usepackage{cleveref}\n\\crefname{section}{§}{§§}\n\\Crefname{section}{§}{§§}\n" +
@@ -164,14 +155,6 @@ describe("paper/section-word", () => {
       "\\documentclass{article}\n\\crefname{section}{§}{§§}\n\\begin{document}\nSee §5.\n\\end{document}\n";
     expect(ids(await lint(src), "paper/section-word")).toHaveLength(1);
   });
-
-  it("markdown: `§5` → `Section 5`, code untouched", async () => {
-    const src = "See §5 and `§3` here.\n";
-    expect(ids(await lint(src, "paper.md"), "paper/section-word")).toHaveLength(
-      1,
-    );
-    expect(await fixed(src, "paper.md")).toBe("See Section 5 and `§3` here.\n");
-  });
 });
 
 describe("paper/leading-zero", () => {
@@ -237,17 +220,6 @@ describe("paper/leading-zero", () => {
       doc("At $p < 0.05$ and $d=0.21$; cell & 0.037 \\\\ and .5 or 0.25."),
     );
   });
-
-  it("markdown: prose and a table cell are fixed, code is not", async () => {
-    const src =
-      "Significant at p<.05.\n\n| task | p |\n| --- | --- |\n| bugfix | .002 |\n\nRun `.25`.\n";
-    expect(ids(await lint(src, "paper.md"), "paper/leading-zero")).toHaveLength(
-      2,
-    );
-    expect(await fixed(src, "paper.md")).toBe(
-      "Significant at p<0.05.\n\n| task | p |\n| --- | --- |\n| bugfix | 0.002 |\n\nRun `.25`.\n",
-    );
-  });
 });
 
 describe("a finding after a heading written over several lines", () => {
@@ -290,15 +262,11 @@ describe("paper/leading-zero — a number after ¶ or § is not a decimal (#102)
     expect(await fixed(doc(CITE))).toBe(doc(CITE));
   });
 
-  it("…while a real decimal still fixes, in markdown too, even beside a cited paragraph", async () => {
+  it("…while a real decimal still fixes, even beside a cited paragraph", async () => {
     const body = `${CITE}\n\nWe use a threshold of .05 throughout.`;
     expect(await fixed(doc(body))).toBe(
       doc(`${CITE}\n\nWe use a threshold of 0.05 throughout.`),
     );
-    const md = "See ¶¶.42 and .44.\n\nSignificant at p<.05.\n";
-    expect(await fixed(md, "paper.md")).toBe(
-      "See ¶¶.42 and .44.\n\nSignificant at p<0.05.\n",
-    );
   });
 });
 
@@ -330,69 +298,7 @@ describe("paper/figure-ref-style", () => {
   });
 });
 
-describe("bib/reachable-entry", () => {
-  const bibDoc = (entries) =>
-    `\\documentclass{acmart}\n\\begin{filecontents*}{refs.bib}\n${entries}\n\\end{filecontents*}\n\\begin{document}\nx\n\\end{document}\n`;
-
-  it("a doi, a url or an arXiv id each make an entry reachable", async () => {
-    const res = await lint(
-      bibDoc(
-        "@article{a, doi = {10.1/x}}\n@misc{b, url = {https://x.org}}\n@misc{c, note = {arXiv:2310.05736}}\n@string{v = {x}}",
-      ),
-    );
-    expect(ids(res, "bib/reachable-entry")).toEqual([]);
-  });
-
-  it("reports on the entry itself, naming its key", async () => {
-    const res = await lint(
-      bibDoc("@misc{ok, url = {https://x.org}}\n@article{lost, title = {x}}"),
-    );
-    const [m] = ids(res, "bib/reachable-entry");
-    expect(m.line).toBe(4);
-    expect(m.column).toBe(1);
-    expect(m.message).toMatch(/`lost`/);
-  });
-
-  it("an entry with no key is reported as `?`, not skipped", async () => {
-    const [m] = ids(
-      await lint(bibDoc("@article{\n  title = {x}}")),
-      "bib/reachable-entry",
-    );
-    expect(m.message).toMatch(/`\?`/);
-  });
-
-  it("a source code with only `text`, or with neither text nor raw", () => {
-    const run = (sourceCode) => {
-      const out = [];
-      bib.rules["reachable-entry"]
-        .create({
-          sourceCode: {
-            getLocFromIndex: (i) => ({ line: 1, column: i }),
-            ...sourceCode,
-          },
-          report: (r) => out.push(r.data.key),
-        })
-        ["root:exit"]();
-      return out;
-    };
-    expect(run({ text: bibDoc("@article{t, title = {x}}") })).toEqual(["t"]);
-    expect(run({})).toEqual([]);
-  });
-});
-
 describe("the escape hatch — a disable directive in a `%` comment", () => {
-  it("🔴 inside the filecontents bibliography, it silences exactly the next entry", async () => {
-    const res = await lint(
-      `\\documentclass{acmart}\n\\begin{filecontents*}{refs.bib}\n` +
-        "% eslint-disable-next-line bib/reachable-entry -- an invited talk, no recording exists\n" +
-        "@misc{talk, title = {x}}\n@misc{lost, title = {y}}\n" +
-        "\\end{filecontents*}\n\\begin{document}\nx\n\\end{document}\n",
-    );
-    const found = ids(res, "bib/reachable-entry");
-    expect(found).toHaveLength(1);
-    expect(found[0].message).toMatch(/`lost`/);
-  });
-
   it("in the body: next-line, same-line, and a disable/enable region", async () => {
     const src = doc(
       [
@@ -467,48 +373,6 @@ describe("where visible text comes from — the walk's other paths", () => {
   });
 });
 
-describe("markdown: offsets that cannot be followed, and rules that are LaTeX-only", () => {
-  const mdLint = async (text, rules) => {
-    const [res] = await new ESLint({
-      cwd: FIX,
-      overrideConfigFile: true,
-      overrideConfig: [
-        {
-          files: ["**/*.md"],
-          plugins: { markdown, paper: typography },
-          language: "markdown/gfm",
-          rules,
-        },
-      ],
-    }).lintText(text, { filePath: join(FIX, "inline", "paper.md") });
-    expect(res.messages.filter((m) => m.fatal)).toEqual([]);
-    return res;
-  };
-
-  it("`§` before prose is reported without a fix", async () => {
-    const src = "As § above.\n";
-    const res = await mdLint(src, { "paper/section-word": "warn" });
-    expect(res.messages.map((m) => [m.ruleId, m.column, m.fix])).toEqual([
-      ["paper/section-word", 4, undefined],
-    ]);
-  });
-
-  it("a sign or a dot written as a character reference has no source offset: nothing is reported", async () => {
-    const res = await mdLint("See &#167; 5 and &#46;25.\n", {
-      "paper/section-word": "warn",
-      "paper/leading-zero": "warn",
-    });
-    expect(res.messages).toEqual([]);
-  });
-
-  it("figure-ref-style is LaTeX-only: silent on markdown", async () => {
-    const res = await mdLint("Figure~\\ref{a} and Fig.~\\ref{b}.\n", {
-      "paper/figure-ref-style": "warn",
-    });
-    expect(res.messages).toEqual([]);
-  });
-});
-
 describe("a source code the rules did not expect — the defaults hold", () => {
   const reports = (ruleId, sourceCode) => {
     const out = [];
@@ -530,12 +394,20 @@ describe("a source code the rules did not expect — the defaults hold", () => {
     ]);
   });
 
-  it("markdown text nodes without positions: seen, never reported at a guessed place", () => {
-    const ast = {
-      type: "root",
-      children: [{ type: "text", value: "§5 and .25" }],
-    };
-    expect(reports("section-word", { text: "§5 and .25", ast })).toEqual([]);
-    expect(reports("leading-zero", { text: "§5 and .25", ast })).toEqual([]);
-  });
+  it.each(["section-word", "leading-zero", "figure-ref-style"])(
+    "%s reads LaTeX only: a source without `raw` (any other language) is not reported",
+    (ruleId) => {
+      const ast = {
+        type: "root",
+        children: [
+          {
+            type: "text",
+            value: "§5 and .25",
+            position: { start: { offset: 0 }, end: { offset: 10 } },
+          },
+        ],
+      };
+      expect(reports(ruleId, { text: "§5 and .25", ast })).toEqual([]);
+    },
+  );
 });

package/eslint-rules/paper-typography.ts

@@ -7,8 +7,8 @@
  *   paper/figure-ref-style  `Fig.~\ref` beside `Figure~\ref` in one document → the majority form
  *
  * The fourth check of the old `paper/typography`, an unreachable bibliography entry, is
- * `bib/reachable-entry` (bib-reachable-entry.mjs): it is about the bibliography, not the prose,
- * and it has no fix.
+ * `bib/reachable-entry` (src/bib-rules.ts): it is about the bibliography, not the prose, and it has
+ * no fix.
  *
  * ── PROVENANCE: every one is a real review finding, not an invented style ───────────────
  * From HotCRP #20 (2026-08-23). Reviewer B listed two as literal to-dos:
@@ -41,11 +41,9 @@
  * IEEE Editorial Style Manual ("0.25, not .25") require the zero. A regex over the raw source
  * found 22 decimals on a real corpus and none was a defect: every one sat in markup (an option,
  * a column spec, a comment, a listing, a tikz coordinate). So this walks the TREE and keeps only
- * what the reader sees as a number:
- *   LaTeX     prose, inline and display math, table cells, and the text arguments of a known
- *             set of text macros. Not: any other macro's arguments, environment arguments,
- *             comments, verbatim/listings/`\lstinline`, tikz, the preamble, the bibliography.
- *   Markdown  text nodes, table cells included. Not: code, inline code, html, front matter.
+ * what the reader sees as a number: prose, inline and display math, table cells, and the text
+ * arguments of a known set of text macros. Not: any other macro's arguments, environment
+ * arguments, comments, verbatim/listings/`\lstinline`, tikz, the preamble, the bibliography.
  * Each character of a run keeps its source offset, so the finding points at the dot and the fix
  * inserts the zero there. The arXiv lookbehind stays: in `2310.05736` the dot follows a digit.
  * A text macro outside the known set (`\hl{.05}`) is read as a parameter and NOT reported — a
@@ -62,7 +60,6 @@
 import assert from "node:assert/strict";
 import { getParser } from "@unified-latex/unified-latex-util-parse";
 import { takeWhile } from "remeda";
-import type { Nodes } from "mdast";
 import {
   texToMdast,
   type TexArg,
@@ -78,7 +75,7 @@ import type {
 } from "./rule-context.ts";
 
 /** Where a `.bib` embedded in a `filecontents` environment sits in the file. */
-export interface BibRange {
+interface BibRange {
   readonly start: number;
   readonly end: number;
   readonly bodyStart: number;
@@ -94,10 +91,17 @@ interface Run {
   offs: (number | null)[];
 }
 
-/** The inline bibliography — a `filecontents` block writing a `.bib` — with its offsets. */
-export function bibRange(text: string): BibRange | null {
+/**
+ * The inline bibliography — a `filecontents` block writing a `.bib`, its name on the `\begin` line or
+ * after white space on a later one, as TeX takes it — with its offsets: the span the markup rules
+ * skip. Which bibliography TeX reads is the build's record (`_build/sources.json`,
+ * docs/design/paper-sources.md §1);
+ * this range only says which bytes are not prose, and moves onto the one projection of live text
+ * with the rest of `skippedRanges` (Q3, docs/design/paper-sources.md §10).
+ */
+function bibRange(text: string): BibRange | null {
   const m =
-    /\\begin\{filecontents\*?\}(?:\[[^\]]*\])?\{[^}]*\.bib\}\r?\n([\s\S]*?)\\end\{filecontents\*?\}/d.exec(
+    /\\begin\{filecontents\*?\}\s*(?:\[[^\]]*\]\s*)?\{[^}]*\.bib\}[^\n]*\n([\s\S]*?)\\end\{filecontents\*?\}/d.exec(
       text,
     );
   if (!m) return null;
@@ -108,7 +112,7 @@ export function bibRange(text: string): BibRange | null {
   return { start: m.index, end: m.index + whole.length, bodyStart, body };
 }
 
-/** A node of a file's projection, as far as its place goes: mdast's, or the `.tex` language's. */
+/** A node of the `.tex` language's projection, as far as its place goes. */
 interface ProjectedNode {
   readonly type: string;
   readonly position?:
@@ -372,66 +376,6 @@ export function texVisibleRuns(
   return texRuns(document ? document.content : root, false, raw, []);
 }
 
-// Markdown node types that are not prose.
-const MD_HIDDEN = new Set([
-  "code",
-  "inlineCode",
-  "html",
-  "yaml",
-  "toml",
-  "math",
-  "inlineMath",
-]);
-/**
- * The source offset of each character of a text node's value. The value is the source with
- * markdown's escapes (`\*`) and character references (`<`) resolved, so the two are walked
- * side by side; a character whose source cannot be followed gets null (reported, not fixed).
- */
-function mdOffsets(value: string, src: string, start: number) {
-  const offs: (number | null)[] = [];
-  let j = start;
-  for (const ch of value) {
-    if (src[j] === ch) offs.push(j++);
-    else if (src[j] === "\\" && src[j + 1] === ch) {
-      offs.push(j + 1);
-      j += 2;
-    } else if (src[j] === "&" && /^&#?\w{1,8};/.test(src.slice(j, j + 10))) {
-      offs.push(null);
-      j = src.indexOf(";", j) + 1;
-    } else {
-      offs.push(null);
-      j++;
-    }
-  }
-  return offs;
-}
-
-function mdVisibleRuns(
-  node: Nodes | undefined,
-  src: string,
-  out: Run[] = [],
-): Run[] {
-  if (!node || MD_HIDDEN.has(node.type)) return out;
-  if (node.type === "text") {
-    const start = node.position?.start.offset;
-    out.push({
-      text: node.value,
-      offs:
-        typeof start === "number"
-          ? mdOffsets(node.value, src, start)
-          : Array.from(node.value, () => null),
-    });
-  }
-  for (const c of "children" in node ? node.children : [])
-    mdVisibleRuns(c, src, out);
-  return out;
-}
-
-const visibleRuns = (sourceCode: RuleSourceCode): Run[] =>
-  isTex(sourceCode)
-    ? texVisibleRuns(sourceCode.raw)
-    : mdVisibleRuns(sourceCode.ast, sourceCode.text);
-
 /** A report at `[from, to)` of the file. */
 const at = (
   context: RuleContext,
@@ -455,36 +399,22 @@ const SECTION_GLYPH = /§/g;
 
 function sectionWord(context: RuleContext) {
   const sc = context.sourceCode;
-  if (isTex(sc)) {
-    const seen = new Set();
-    for (const m of markupMatches(sc, SECTION_TEX)) {
-      seen.add(m.index);
-      const word = m[3] ? "Section~" : "Section ";
-      const range: [number, number] = [m.index, m.index + m[0].length];
-      at(context, range[0], range[1], {
-        messageId: "sign",
-        fix: (f) => f.replaceTextRange(range, word),
-      });
-    }
-    // A glyph before neither a reference nor a number: reported, not rewritten — there is no
-    // single right word for a doubled sign or a sign before prose.
-    for (const m of markupMatches(sc, SECTION_GLYPH))
-      if (!seen.has(m.index))
-        at(context, m.index, m.index + 1, { messageId: "sign" });
-    return;
+  if (!isTex(sc)) return;
+  const seen = new Set();
+  for (const m of markupMatches(sc, SECTION_TEX)) {
+    seen.add(m.index);
+    const word = m[3] ? "Section~" : "Section ";
+    const range: [number, number] = [m.index, m.index + m[0].length];
+    at(context, range[0], range[1], {
+      messageId: "sign",
+      fix: (f) => f.replaceTextRange(range, word),
+    });
   }
-  for (const r of visibleRuns(sc))
-    for (const m of r.text.matchAll(/§([ \t]*)(\d)?/g)) {
-      const from = r.offs[m.index];
-      if (from === null || from === undefined) continue;
-      // The sign and the space after it; the number stays.
-      const [, gap = ""] = m;
-      const range: [number, number] = [from, from + 1 + gap.length];
-      at(context, range[0], range[1], {
-        messageId: "sign",
-        ...(m[2] ? { fix: (f) => f.replaceTextRange(range, "Section ") } : {}),
-      });
-    }
+  // A glyph before neither a reference nor a number: reported, not rewritten — there is no
+  // single right word for a doubled sign or a sign before prose.
+  for (const m of markupMatches(sc, SECTION_GLYPH))
+    if (!seen.has(m.index))
+      at(context, m.index, m.index + 1, { messageId: "sign" });
 }
 
 // ── paper/leading-zero ─────────────────────────────────────────────────────────────────
@@ -508,8 +438,9 @@ const paragraphBefore = (src: string, i: number) => {
 
 function leadingZero(context: RuleContext) {
   const sc = context.sourceCode;
-  const src = isTex(sc) ? sc.raw : sc.text;
-  for (const r of visibleRuns(sc))
+  if (!isTex(sc)) return;
+  const src = sc.raw;
+  for (const r of texVisibleRuns(src))
     for (const m of r.text.matchAll(BARE_DECIMAL)) {
       const dot = r.offs[m.index];
       if (dot === null || dot === undefined) continue;

package/eslint-rules/papers.harness.mjs

@@ -170,7 +170,7 @@ for (const bad of ["", null, 0, false, [], {}, 42])
   const names = Object.keys(sets).sort();
   assert.deepEqual(
     names,
-    ["all", "md", "pdfFacts", "refFacts", "status", "tex", "venue"],
+    ["all", "pdfFacts", "refFacts", "status", "tex", "venue"],
     "the glob set changed — a consumer's config enumerates these by name, and a renamed or " +
       "dropped key is a block that silently lints nothing",
   );

package/eslint-rules/papers.ts

@@ -41,7 +41,7 @@
  *
  *   A FILENAME MARKER, no root at all (`**\/paper.tex`).
  *     Measured on the first consumer: `paper.tex` ×4 with one outside the corpus,
- *     `PIPELINE-STATUS.md` ×6 with two of them test fixtures, `paper.md` ×3 with two outside.
+ *     `PIPELINE-STATUS.md` ×6 with two of them test fixtures.
  *     The guard that would consume this is a `deny` at `error` severity; a denial on writing
  *     a fixture gets the hook switched off the same day.
  *
@@ -134,15 +134,13 @@ export function papersRoot(
  * The glob sets a consumer declares its blocks on, derived from one root.
  *
  * 🔴 THE SHAPES LIVE HERE, NOT IN THE CONSUMER, and that is the half that keeps a rule from
- * going blind later. `paper.md` + `draft.md`, `_build/*.facts.json`, `PIPELINE-STATUS.md` are
+ * going blind later. `paper.tex`, `_build/*.facts.json`, `PIPELINE-STATUS.md` are
  * this pipeline's CONVENTION; if a rule starts reading a new file, the glob for it must arrive
  * with the rule, in the same package, rather than as a line every consumer has to add by hand
  * to keep up. A hand-maintained list of paths is the thing that rots, and rots silently.
  */
 export function paperFiles(root: string) {
   return {
-    /** Markdown drafts — the two names a paper's prose is allowed to have. */
-    md: [`${root}/*/paper.md`, `${root}/*/draft.md`],
     /** The LaTeX source of a paper (not its figures, not its vendored drafts). */
     tex: [`${root}/*/paper.tex`],
     /** The per-paper stage ledger. */

package/eslint-rules/tex-build.harness.mjs

@@ -35,8 +35,8 @@
  *   (`284:will be released`, `284:Upon acceptance` + `285:will be made publicly available`).
  *
  * 🔴 AND ONE DISAGREEMENT, WHICH IS NOT IN THE LOGIC BUT IN THE INPUT — and it is the whole
- * gain. The old implementation picked its file by looking for `paper.md` / `draft.md` and
- * otherwise taking the FIRST `.tex` in `readdirSync` order. Measured 2026-09-07:
+ * gain. The old implementation picked its file by taking the FIRST `.tex` in `readdirSync`
+ * order. Measured 2026-09-07:
  *   paper A → a 536-BYTE file of nine `\def`s with venue numbers. The check, added
  *             specifically BECAUSE of paper A, was not reading paper A;
  *   paper B → a four-month-old draft, one of seven `.tex` files in that directory;

package/eslint-rules/tex-build.ts

@@ -3,27 +3,11 @@
  * how the prose is written.
  *
  * ── WHY ITS OWN NAMESPACE, AND NOT `paper/*` ────────────────────────────────
- * In the source corpus the `paper/*` family is declared on BOTH `paper.md` and `paper.tex`,
- * and that parity is pinned by an assertion: "a rule added for markdown must not silently
- * fail to reach LaTeX (or the other way round)". The subject here is the BUILD MODE
- * (`\documentclass[review]`, `printacmref=false`), and in markdown it does not exist: a draft
- * has no mechanical sign saying "this build is the final one". Writing
- * `paper/future-promise: "off"` on markdown would be an exemption justified by "there is no
- * subject", and that same parity assertion requires the rule to be ALIVE on the other side —
- * i.e. the machinery exists but is built for the opposite direction (off on `.tex`, alive on
- * `.md`).
- *
- * Hence a namespace of its own, the same way an external word-counter gets one: it has a word
- * count markdown does not have, we have a preamble markdown does not have. Where the config
- * enumerates non-`paper/*` rules on `.tex`, the list is CLOSED and `tex/future-promise` is
- * named explicitly, so that a new namespace arrives by decision rather than as a side effect
- * of editing a config.
- *
- * ⚠️ THE MEASUREMENT THIS DECISION RESTS ON (source corpus, 2026-09-06 → 2026-09-07): had the
- * rule been declared on `paper.md` it would produce ZERO there TODAY — not one match of
- * `FUTURE_PROMISE_RE` in the markdown draft of the one paper that has one. So the choice is
- * about semantics, not noise: markdown has no input distinguishing a draft from a shipped
- * build, and the rule would fire on a draft's honest promise.
+ * The subject here is the BUILD MODE (`\documentclass[review]`, `printacmref=false`), not the
+ * prose the `paper/*` rules read. Hence a namespace of its own, the same way an external
+ * word-counter gets one. Where the config enumerates non-`paper/*` rules on `.tex`, the list is
+ * CLOSED and `tex/future-promise` is named explicitly, so that a new namespace arrives by
+ * decision rather than as a side effect of editing a config.
  *
  * ── THE RULE READS THE FILE FROM DISK, NOT THE PROJECTION. This is load-bearing ──
  * The LaTeX language in `latex-language.mjs` hands rules a PROJECTION of `.tex` into a
@@ -37,9 +21,8 @@
  * run (and in CI) there is no such difference.
  *
  * ── WHAT THE MOVE TO A LINT RULE FIXED BESIDES THE FORM (measured 2026-09-07) ──
- * 🔴 The previous implementation, a script, picked its input like this: look for `paper.md` /
- * `draft.md`, and failing that take `readdirSync(dir).find((f) => f.endsWith(".tex"))` — the
- * FIRST `.tex` in directory order. Measured across the three papers of the source corpus:
+ * 🔴 The previous implementation, a script, picked its input like this:
+ * `readdirSync(dir).find((f) => f.endsWith(".tex"))` — the FIRST `.tex` in directory order. Measured across the three papers of the source corpus:
  *   paper A → a 536-BYTE file holding nine `\def`s with venue numbers. The check, added
  *             specifically BECAUSE of paper A, was not reading paper A;
  *   paper B → a four-month-old draft, one of SEVEN `.tex` files in that directory (the real
@@ -121,7 +104,7 @@ const FRONTMATTER_OVERRIDES = [
 // is looking.
 //
 // The scope is DELIBERATELY NARROW, because a noisy gate is a silenced gate:
-//   - `.tex` only (see the file header: markdown has no sign of "this build is final");
+//   - `.tex` only: the build mode is read from the preamble;
 //   - a review-mode build is EXEMPT, and that is the whole point: during peer review
 //     "at camera-ready" is a true and ordinary promise. The defect is only a promise that
 //     outlived its own delivery;

package/eslint.config.mjs

@@ -520,7 +520,6 @@ const TYPESCRIPT_ONLY = [
   "test/e2e/**/*.{js,mjs,cjs}",
   "test/fixtures/layers/src/**/*.{js,mjs,cjs}",
   "fixtures/accepted-papers/**/*.{js,mjs,cjs}",
-  "fixtures/real-markdown-paper/**/*.{js,mjs,cjs}",
   "skills/paper-pipeline/scripts/fixtures/**/*.{js,mjs,cjs}",
   "skills/plan-paper-timeline/fixtures/**/*.{js,mjs,cjs}",
 ];
@@ -799,6 +798,28 @@ export default [
       "no-restricted-imports": ["error", { paths: ONE_COLLECTION_LIBRARY }],
     },
   },
+  // 🔴 ONE PLACE READS BIBLIOGRAPHIES. `@retorquere/bibtex-parser` is known to its adapter only;
+  // every other file — the app, the domain, another adapter, a skill's script — reads a `.bib` through
+  // the `BibReader` port, and a path a person gave through `bibliographyAt` (src/paper-sources.ts).
+  // Three readers of `.bib` text grew here before there was one. Its own rule id, so no other
+  // `no-restricted-imports` block can replace it (test/eslint-bib-parser.test.ts).
+  {
+    files: ["**/*.{ts,mts,mjs,js,cjs}"],
+    ignores: ["src/adapters/bibtex/**"],
+    plugins: { n },
+    rules: {
+      "n/no-restricted-import": [
+        "error",
+        [
+          {
+            name: ["@retorquere/bibtex-parser", "@retorquere/bibtex-parser/**"],
+            message:
+              "The bibtex parser is src/adapters/bibtex/'s: read a .bib through the BibReader port (src/ports/bib-reader.ts), or a path through bibliographyAt (src/paper-sources.ts).",
+          },
+        ],
+      ],
+    },
+  },
   // 🔴 THE LINTER DOES NOT READ A SKILL. Skills are consumers of the package, not its storage: the
   // venue presets once lived in a skill and were located through a skill's script, so renaming the
   // skill broke `paperlint lint` (#133). The package's own code — src/, eslint-rules/, lib/ — may

package/fixtures/build-e2e/aidc/paper.tex

@@ -3,7 +3,7 @@
 % rule is judged against what a real build of the class measures. It conforms: the LLM Usage
 % Statement closes it, and not one rule has anything to say.
 \documentclass[conference,compsoc]{IEEEtran}
-\begin{filecontents*}{refs.bib}
+\begin{filecontents*}[overwrite]{refs.bib}
 @article{knuth84,
   title   = {Literate Programming},
   author  = {Knuth, Donald E.},

package/fixtures/build-e2e/anon-ack/paper.tex

@@ -1,7 +1,7 @@
 % B: the name in an acknowledgments paragraph.
 % The declared identity is invented: "Ada Example", handle "adaexample" (paperlint.json).
 \documentclass[conference,compsoc]{IEEEtran}
-\begin{filecontents*}{refs.bib}
+\begin{filecontents*}[overwrite]{refs.bib}
 @article{knuth84,
   title   = {Literate Programming},
   author  = {Knuth, Donald E.},

package/fixtures/build-e2e/anon-author/paper.tex

@@ -1,7 +1,7 @@
 % A: the name in the author block.
 % The declared identity is invented: "Ada Example", handle "adaexample" (paperlint.json).
 \documentclass[conference,compsoc]{IEEEtran}
-\begin{filecontents*}{refs.bib}
+\begin{filecontents*}[overwrite]{refs.bib}
 @article{knuth84,
   title   = {Literate Programming},
   author  = {Knuth, Donald E.},