paperlint 7.1.0 → 8.0.0

package/src/venue-rules.ts

@@ -37,6 +37,8 @@
  *   no geometry (no banal)  pdf/measured (warn); geometry and body-size silent, page-limit judges
  *                           only a body counted before the references; fonts still judge
  *   kind does not resolve   pdf/profile (error); page-limit silent; everything else still judges
+ *   cycle phase `porting`   every rule silent — the source is in the previous venue's template by
+ *                           declaration (src/domain/cycle.ts); cycle/port-done watches the declaration
  */
 import { dirname, isAbsolute, join, basename, relative } from "node:path";
 import {
@@ -65,6 +67,7 @@ import {
   type PageSplit,
 } from "./domain/body-pages.ts";
 import type { TextFacts } from "./facts-file.ts";
+import { effectiveBlind, isPorting } from "./domain/cycle.ts";
 
 // ── the verdict's vocabulary ─────────────────────────────────────────────────────────
 
@@ -96,8 +99,19 @@ export interface Resolved {
 export type Assessment =
   | { readonly kind: "no-venue" }
   /** A `paperlint.json` that extends no preset yet — `pdf/measured` names the file to set. */
-  | { readonly kind: "no-preset"; readonly file: string }
+  | {
+      readonly kind: "no-preset";
+      readonly file: string;
+      /** The paper keeps `cycles`: its venue is the current cycle's, and `extends` beside them is refused. */
+      readonly cycles: boolean;
+    }
   | { readonly kind: "unresolved"; readonly finding: Finding }
+  /**
+   * The current cycle declares its port to the venue's template as open work (`"phase": "porting"`):
+   * a PDF built with the previous venue's template says nothing about this venue's limits, so every
+   * venue rule is silent, and `cycle/port-done` speaks when the port is done and the phase still says so.
+   */
+  | { readonly kind: "porting" }
   | { readonly kind: "unbuilt"; readonly venue: Resolved }
   | { readonly kind: "stale"; readonly finding: Finding }
   | {
@@ -147,9 +161,13 @@ function kindOf(
 /** How the venue reviews, and who the paper says wrote it: what `anonymity/*` judges by. */
 const reviewOf = (
   preset: Pick<Preset, "blind">,
-  settings: Pick<PaperSettings, "identity">,
+  settings: Pick<PaperSettings, "identity" | "cycles">,
 ): Pick<Resolved, "blind" | "identity"> => ({
-  blind: preset.blind,
+  // An accepted attempt's camera-ready carries the authors: no longer blind.
+  blind: effectiveBlind(
+    preset.blind,
+    settings.cycles?.current ?? { kind: "none" },
+  ),
   identity: settings.identity,
 });
 
@@ -174,13 +192,25 @@ function freshFacts(
 
 const isFinding = (v: object): v is Finding => "messageId" in v;
 
+/** The paper's declared venue and kind, resolved: what the judges compare the facts against. */
+const resolvedOf = (preset: Preset, settings: PaperSettings): Resolved => ({
+  venue: preset.label,
+  format: preset.format,
+  ...kindOf(preset.label, preset.format, settings.kind),
+  ...reviewOf(preset, settings),
+});
+
 /** One paper, assessed. Reads through `deps.files` only; never throws on a paper's files. */
 export function assessPaper(paperDir: string, deps: VenueRuleDeps): Assessment {
   const p = paperPreset(paperDir, deps);
   if (p.kind === "none")
     return p.settings === null
       ? { kind: "no-venue" }
-      : { kind: "no-preset", file: join(paperDir, CONFIG_FILE) };
+      : {
+          kind: "no-preset",
+          file: join(paperDir, CONFIG_FILE),
+          cycles: p.settings.cycles !== null,
+        };
   if (p.kind === "settings-problem")
     return {
       kind: "unresolved",
@@ -191,13 +221,9 @@ export function assessPaper(paperDir: string, deps: VenueRuleDeps): Assessment {
       kind: "unresolved",
       finding: finding("preset", { why: presetProblemText(p.problem) }),
     };
-  const label = p.preset.label;
-  const venue: Resolved = {
-    venue: label,
-    format: p.preset.format,
-    ...kindOf(label, p.preset.format, p.settings.kind),
-    ...reviewOf(p.preset, p.settings),
-  };
+  if (p.settings.cycles !== null && isPorting(p.settings.cycles.current))
+    return { kind: "porting" };
+  const venue = resolvedOf(p.preset, p.settings);
   const factsText = text(deps.files, factsPath(paperDir));
   if (factsText === null) return { kind: "unbuilt", venue };
   const facts = freshFacts(deps.files, paperDir, factsText);
@@ -558,7 +584,11 @@ const JUDGES: Readonly<Record<VenueRuleName, Judge>> = {
   fresh: (a) => (a.kind === "stale" ? [a.finding] : []),
   measured: (a, { shown }) =>
     a.kind === "no-preset"
-      ? [finding("noPreset", { file: shown(a.file) })]
+      ? [
+          finding(a.cycles ? "noPresetCycle" : "noPreset", {
+            file: shown(a.file),
+          }),
+        ]
       : a.kind === "unbuilt"
         ? [
             finding("unbuilt", {
@@ -655,6 +685,8 @@ const META: Readonly<Record<VenueRuleName, Meta>> = {
         "not built yet, so `{{venue}}`'s page limit, fonts and format are not checked — run `paperlint build`",
       noPreset:
         'this paper names no venue preset yet, so its page limit, fonts and format are not checked — set "extends" in {{file}} (e.g. "paperlint:agenticdev"; see docs/rules.md)',
+      noPresetCycle:
+        'this paper\'s current cycle names no venue preset, so its page limit, fonts and format are not checked — set the venue in the current cycle in {{file}} ("venue": { "kind": "preset", "extends": "paperlint:…" }), or open a new cycle when the last one has ended',
       noGeometry:
         "the build measured no page geometry (banal was not found or failed), so page size, columns, page limits and font sizes were NOT checked — run `paperlint toolchain`, then `paperlint build`",
     },

package/fixtures/real-markdown-paper/baseline.json

@@ -1,22 +0,0 @@
-{
-  "$comment": [
-    "What the rules say about this article TODAY, as data rather than as a sentence in a docblock.",
-    "It is not a target and not a clean bill of health — the fixture's README says findings here",
-    "are measurements. Its job is the SILENCE half: a rule that starts reporting something new on",
-    "225 lines of real prose has almost certainly gained a false positive, and the harness beside",
-    "this file names the rule. That is not expressible on an eleven-line stub, where every line was",
-    "written by someone who already knew which rule would read it.",
-    "",
-    "Recorded 2026-09-19, re-recorded 2026-09-26 for 3.0.0. Two of the three rules are about the",
-    "scorecard being deliberately thin (no pdf, no frozen source) rather than about the prose.",
-    "paper/leading-zero reports each of the 18 p-values of paperlint#44 where it is (one finding per",
-    "occurrence since 3.0.0; before, paper/typography reported them as one count). They are real",
-    "findings against IEEE / ISO 80000-1 style, in a blog post written for a general audience.",
-    "The count is expected to stay. A drop is never a failure here, only growth is."
-  ],
-  "findings": {
-    "paper/stages": 2,
-    "paper/source": 1,
-    "paper/leading-zero": 18
-  }
-}

package/fixtures/real-markdown-paper/baseline.ts

@@ -1,70 +0,0 @@
-/**
- * ONE reader of `baseline.json`, shared by every run that compares against one: the harness beside
- * this file (the repository's own `bin/paperlint.mjs`), `test/e2e/install/install.e2e.ts` (the
- * binary a consumer actually got), and `fixtures/accepted-papers/accepted-papers.test.ts` (one
- * baseline per accepted paper). Two copies of "growth fails, a drop never does" would drift the
- * first time one of them is tightened, and the runs would then disagree about the same rules.
- */
-import { readFileSync } from "node:fs";
-import { dirname, join } from "node:path";
-import { fileURLToPath } from "node:url";
-import { z } from "zod";
-
-/** Findings per rule id. */
-export type Counts = Record<string, number>;
-
-const Baseline = z.object({ findings: z.record(z.string(), z.number()) });
-const LintOutput = z.array(
-  z.object({
-    messages: z.array(z.object({ ruleId: z.string().nullable() })).nullish(),
-  }),
-);
-
-const HERE = dirname(fileURLToPath(import.meta.url));
-
-/** The counts recorded in `dir`'s `baseline.json` (this article's by default), `{ruleId: n}`. */
-export function recordedFindings(dir: string = HERE): Counts {
-  return Baseline.parse(
-    JSON.parse(readFileSync(join(dir, "baseline.json"), "utf8")),
-  ).findings;
-}
-
-/**
- * `paperlint lint --json` stdout → `{ruleId: n}`.
- * 🔴 Output that does not parse THROWS. An empty set read as "clean" is how this repository's
- * checks have gone hollow before.
- */
-export function countByRule(stdout: string): Counts {
-  const out: Counts = {};
-  for (const file of LintOutput.parse(JSON.parse(stdout)))
-    for (const m of file.messages ?? []) {
-      const rule = String(m.ruleId);
-      out[rule] = (out[rule] ?? 0) + 1;
-    }
-  return out;
-}
-
-/**
- * Compare measured counts with the recording.
- *   grew     — a rule says MORE about real prose than recorded: a false positive until shown
- *              otherwise. Always a failure.
- *   vanished — a recorded rule went fully quiet without the recording being updated: how a check
- *              dies unnoticed. Also a failure.
- * A partial drop is neither: it is what a fix looks like (paperlint#44 is expected to take typography
- * to zero, and then the recording is updated in the same change).
- */
-export function compareToBaseline(
-  found: Readonly<Counts>,
-  recorded: Readonly<Counts> = recordedFindings(),
-): {
-  grew: { rule: string; now: number; recorded: number }[];
-  vanished: string[];
-} {
-  const grew = Object.entries(found)
-    .filter(([rule, n]) => n > (recorded[rule] ?? 0))
-    .map(([rule, n]) => ({ rule, now: n, recorded: recorded[rule] ?? 0 }));
-  const vanished = Object.keys(recorded).filter(
-    (r) => (recorded[r] ?? 0) > 0 && !(r in found),
-  );
-  return { grew, vanished };
-}

package/skills/paper-pipeline/PIPELINE-MAP.md

@@ -1,381 +0,0 @@
----
-title: "Paper pipeline map — what runs where"
-created: 2026-08-26
-updated: 2026-08-26
-tags: [paper-pipeline, ci, hooks, map]
----
-
-# What runs where: the pipeline map
-
-Assembled 2026-08-26 **from facts** — `.claude/settings.json` and
-`.github/workflows/paper-gates.yml` — not from memory. Occasion (translated from Russian): "I have
-no fucking idea how the pipeline works right now."
-
-> ⭐ **How to build checks** (the ladder, occupancy across 14 tools, measurements) —
-> in the author's private notes: `<papers-root>/research/2026-08-26-sessiya-arhitektura-payplayna.md`
-
-## Four tiers, and they fire at different times
-
-```
-┌─ TIER 1: WHILE WRITING ────────────────────── instant, in-session ──┐
-│                                                                      │
-│  you edit paper.md / paper.tex                                      │
-│         │                                                            │
-│         ├─ PreToolUse   → paper-lint pre      🛑 THE ONLY BLOCKER   │
-│         │                  paragraph >200 words · ≥9 numbers ·      │
-│         │                  sentence >55 words · refuses the edit    │
-│         │                  outright                                 │
-│         │                                                            │
-│         ├─ PreToolUse   → paper-edit-guard    🛑 a write from Bash  │
-│         │                  (bypassing PostToolUse gates) → deny     │
-│         │                                                            │
-│         └─ PostToolUse  → paper-lint post     ⚠️ nudges only        │
-│                            plain-language · carries: · SHAVE        │
-│                            prose-lint · the numbers gate             │
-│                                                                      │
-│  end of response → Stop → paper-lint stop  (throttled to 1 hour)    │
-│                        15 checks across all papers                   │
-│                        + kb-lint stop: 12 checks across the base    │
-└──────────────────────────────────────────────────────────────────────┘
-                                   │
-                                   ▼
-┌─ TIER 2: BUILD ───────────────────── when you run pdflatex ─────────┐
-│                                                                      │
-│   paper.tex                                                          │
-│      │                                                               │
-│      ▼                                                               │
-│   pdflatex ──┬─→ INSIDE, acmart runs (the document class)            │
-│              │   it's the same thing that complains in the log:      │
-│              │   no CCS · no libertine ·                              │
-│              │   «ACM reference format is mandatory»                 │
-│              │                                                       │
-│              ├─→ paper.pdf   ← the final artifact, fonts EMBEDDED    │
-│              ├─→ paper.log   ← acmart writes its warnings here       │
-│              └─→ paper.blg   ← bibtex writes bibliography errors here│
-│                                                                      │
-│   check-render.sh reads ALL THREE:                                   │
-│      log → overfull · undefined refs · class warnings                │
-│      blg → dropped entries · cited-but-absent                        │
-│      pdf → fonts (LinLibertine/LinBiolinum) ← added 25.08            │
-│      + chktex on the .tex, if installed (advisory)                   │
-└──────────────────────────────────────────────────────────────────────┘
-                                   │
-                                   ▼
-┌─ TIER 3: CI ─────────────── on push/PR, 7 jobs ──────────────────────┐
-│                                                                      │
-│  changes ──┐ a path classifier, ~10s, WITHOUT a checkout             │
-│            │ decides which jobs even run                            │
-│            ▼                                                         │
-│  ┌─────────────────────────────────────────────────────────────┐    │
-│  │ prose      eslint(md) · paper-lint --gate 🛑 · bib-authors  │    │
-│  │            prose-lint · artifact coverage · population map  │    │
-│  │            structure · provenance · verify-cites · scorecard│    │
-│  ├─────────────────────────────────────────────────────────────┤    │
-│  │ numbers    paper_numbers selftest → gate                    │    │
-│  │            artifact/reproduce.py ← wired in 26.08            │    │
-│  ├─────────────────────────────────────────────────────────────┤    │
-│  │ anonymity  check-anon over the bundle + every *.selftest.{sh,mjs} │
-│  ├─────────────────────────────────────────────────────────────┤    │
-│  │ build      texlive/texlive:latest container ← FONTS LIVE HERE│    │
-│  │            aclpubcheck (optional) · builds · check-render    │    │
-│  ├─────────────────────────────────────────────────────────────┤    │
-│  │ hooks      vigiles lint 🛑 · every harness on disk           │    │
-│  │            TeXtidote (pinned jar) · ledger selftest          │    │
-│  ├─────────────────────────────────────────────────────────────┤    │
-│  │ skill-firing   once a week, a real model                    │    │
-│  └─────────────────────────────────────────────────────────────┘    │
-└──────────────────────────────────────────────────────────────────────┘
-                                   │
-                                   ▼
-┌─ TIER 4: THE VENUE ──────── the one verdict that actually decides ──┐
-│                                                                      │
-│   eRights (ACM) ──→ issues a block of copyright commands             │
-│         │              ↓ paste into paper.tex, REBUILD               │
-│         ▼                                                            │
-│   upload goes to TWO PLACES:                                         │
-│      • conference-publishing.com  ← goes to typesetting and ACM DL   │
-│      • HotCRP paper/20/edit       ← a copy for the workshop chairs   │
-│         │                                                            │
-│         ▼                                                            │
-│   their checker (TAPS / the portal's format checker) — closed-source │
-└──────────────────────────────────────────────────────────────────────┘
-```
-
-## What only runs by hand
-
-Skills (`grade-paper-writing`, `tighten-paper`, `pc-panel-review`, `cold-read-diff`,
-`verify-citations`, `argument-arc`, `harden-paper`…) are **not invoked by any event**. A human calls
-them. This is the judgment layer; its mechanical parts should keep getting pushed down into tiers
-1–3.
-
-## Where each tier catches the same thing
-
-| defect                                | caught by                                                                    |
-| ------------------------------------- | ---------------------------------------------------------------------------- |
-| a wall-of-text paragraph on insertion | tier 1, blocks                                                               |
-| a number that doesn't match the data  | tier 1 (post) and tier 3 (numbers)                                           |
-| a reference dropped by bibtex         | tier 2 (`.blg`) → tier 3 (build)                                             |
-| **wrong font**                        | tier 2 (the facts' font list) — but it's PREVENTED by tier 3 (the container) |
-| a stage declared with no PDF          | tier 3, `paper-lint --gate` 🛑                                               |
-| ACM format compliance                 | **tier 4, and only tier 4**                                                  |
-
----
-
-# 🗺️ Roadmap: what is NOT built
-
-## 1. 🔴 A requirement as a BUILD FAILURE, not a check afterward (the author's idea, 26.08)
-
-Right now almost everything is caught **after** the build. But `acmart` already knows how to fail
-the build — it does exactly that on missing CCS concepts (_"CCS concepts are mandatory for papers
-over two pages"_). So the mechanism exists, and we're not using it.
-
-**Rung 2 of the ladder instead of rung 3:** there's nothing to check in something that can't even be
-built.
-
-| requirement           | how to turn it into a build failure                                                                              |
-| --------------------- | ---------------------------------------------------------------------------------------------------------------- |
-| **fonts**             | `fontspec` + a font by path → file missing → the build fails instead of silently falling back to Computer Modern |
-| page limit            | `\AtEndDocument` + a check on `\thepage` → `\PackageError`                                                       |
-| text-block dimensions | the class sets these; catch the fact that they were overridden                                                   |
-| no page numbers       | check `\thepage` in the output                                                                                   |
-
-🔴 **MEASURED on 2026-08-26, not assumed: `fontspec` with `pdflatex` DOES NOT WORK.** A minimal
-document, a `pdflatex` run, exit code 1, verbatim from the log:
-
-```
-Package: fontspec  Font selection for XeLaTeX and LuaLaTeX
-! Fatal Package fontspec Error: The fontspec package requires either XeTeX or LuaTeX.
-```
-
-⇒ The whole chain: only `fontspec` can load a font by path → `fontspec` requires XeLaTeX or LuaLaTeX
-→ we run `pdflatex` → **the construction requires SWITCHING THE BUILD ENGINE**, not adding a
-package. That moves the item from "add a line" to "change the program that builds the paper."
-
-⚠️ **Three unverified things that decide whether the first row is even feasible** (these are
-unknowns, not objections): `fontspec` requires XeLaTeX/LuaLaTeX instead of pdflatex · `acmart`
-assigns its own fonts and might conflict · whether the TAPS pipeline will even accept a LuaLaTeX
-source. Plus a risk: switching engines changes how microtype is handled, and there's no page-count
-slack.
-
-**Checked by a cheap spike** — build a copy in scratch, see whether the class fights back and
-whether the layout shifts. Do this AFTER the 29.08 submission.
-
-## 2. Mechanical camera-ready compliance — 1 requirement out of ~25 is checked
-
-ACM's instructions list ~25 mechanical requirements. Checked: fonts. The rest — text-block geometry,
-9pt/8pt type size, column balance, DOIs on references, captions above tables and below figures,
-capitalization, section ordering — **nothing checks any of it**.
-
-**A construction found and verified against source (26.08):**
-
-```
-banal -json  +  pdf.js     →   paper.facts.json   →   @eslint/json
-(geometry)     (fonts)         (plain JSON)           (the official plugin)
-```
-
-- **`banal`** — [kohler/hotcrp/src/banal](https://github.com/kohler/hotcrp/blob/master/src/banal),
-  1901 lines of Perl, dependencies are core Perl + `pdftohtml`. Standalone, alive (commits on
-  03.08.2026). paperlint feeds it pdftohtml-style XML written from pdf.js instead, so poppler is not
-  needed (`src/adapters/banal/`: `xml.ts` writes it, `run.ts` runs banal; issue #61). Gives you `papersize`, `margin`, `bodyfontsize`, `leading`, `columns`, `npages`,
-  `reffontsize`. It **doesn't know about fonts by construction** — the words `type3`/`embed` don't
-  appear in the file at all.
-- **banal does not accept a spec** — the comparison lives in `checkformat.php`, the profile grammar
-  is `papersize;pagelimit;columns;textblock;bodyfontsize;bodylineheight`. For ACM ≈
-  `letter;;2;18x23.5cm;9`.
-- **Why through JSON and not an ESLint language plugin:** `fileType: "binary"` in ESLint is **not
-  implemented**, the docs say verbatim — _"should be 'text' (in the future, we will also support
-  'binary')"_, and the runtime reads the file unconditionally as UTF-8 (`eslint-helpers.js:1281`),
-  irreversibly mangling PDF bytes into U+FFFD. A hack is possible (`parse()` reads the disk itself),
-  but it breaks `--fix`, the cache, and `eslint-disable` — there's nowhere to put a directive inside
-  a PDF.
-- **veraPDF is built the same way** — the one PDF validator with user-defined rules: _"doesn't parse
-  PDF documents directly. Instead it processes the machine readable report output"_. Matching the
-  architecture of an industrial tool is a sign the construction is right.
-
-**Occupancy checked:** `eslint-plugin-pdf` → E404 · `textlint` accepts its own format but requires
-conversion to text first (precedent: `textlint-plugin-pptx`) · `Vale` won't take a custom parser ·
-`MegaLinter`/`super-linter`/`reviewdog` are orchestrators, they won't give us a rule registry for our
-format.
-
-## 3. Venue constants belong in a card, not in code
-
-> **Done (#79, 2.1.0), in a different shape than planned below.** The constants live in **venue
-> presets**, JSONC files in the package's `presets/`: `acm-sigconf.jsonc` is the ACM
-> family (the `format` block — page size, columns, fonts, font sizes — plus the `tex` packages),
-> `agenticdev.jsonc` and `aisec.jsonc` extend it and add their `kinds` (page limits) and `rules`,
-> `realm.jsonc` stands alone. A paper picks one in `<paper>/paperlint.json`
-> (`{ "extends": "paperlint:<venue>", "kind": "…" }`); the `pdf/*` lint rules judge the built PDF
-> against it. The `.md` cards stay as prose. What follows is the plan as it was written.
-
-Right now `LinLibertine`, `LinBiolinum`, `acmart` are hardcoded into `check-render.sh`. This is
-**venue data**, and it belongs in
-`.claude/skills/submit-paper/references/venues/<venue>.md`:
-
-```yaml
-venue:
-  template: acmart
-  page_size: letter
-  fonts: { text: LinLibertine, title: LinBiolinum }
-  textblock: 18x23.5cm
-  body_pt: 9
-  body_limit: 5
-  extra_ref_pages: 2
-```
-
-The second consumer of this block is the comparison against `banal`'s output (six numbers). The
-first is the page limit, which currently exists in **four copies**, and they have already drifted
-apart (`venues/realm.md` says 8, `build-submission.sh` hardcodes 8, `PIPELINE-STATUS` writes "body
-8/8," and the build prints 9).
-
-## 4. `check-render.sh` — 244 lines of bash full of greps
-
-The place is right (nothing else reads the build artifacts, and no offline ACM checker exists),
-**the form isn't**: no check registry, no config, no severity, thresholds are hardcoded. Bolting 25
-more greps onto it is the path we're getting off of. Deal with it together with items 2 and 3.
-
-## 5. Check occupancy BEFORE writing — the rule exists, it's rung 4
-
-Written down twice (05.08, 24.08), and the `sweep-design-space` skill exists. **It was violated
-three times overnight on 25/26.08** — meaning prose and a skill don't work, both are rung 4.
-
-**Proposed construction:** a new check-file must carry an `occupancy:` line — what was searched and
-why it didn't fit; the linter requires its presence. Same form as `carries:` for sections and
-`Cause:` for re-check findings. To be honest: this is rung 3, not 2 — it doesn't stop you from
-skipping the search, but it makes **the absence of evidence visible**.
-
----
-
-# 🧊 FREEZE on the current ad hoc state (the author's decision, 2026-08-26)
-
-Verbatim (translated from Russian): _"mark the current pipeline as shit by rule — don't expand it as
-it is, only allow new, proper stuff."_
-
-> **Fixes — yes. New checks — no.**
-> A new check goes only in the new form — but **only where the new form already exists.**
-
-Without the second half the rule is incoherent, and I slipped on it that same evening: announced the
-freeze and, two messages later, proposed bolting a check onto `paper-lint`.
-
-| check category                  | new form                                | where a new one goes                                           |
-| ------------------------------- | --------------------------------------- | -------------------------------------------------------------- |
-| a single document: markdown, JS | **exists** — ESLint                     | 🛑 not allowed into the old form                               |
-| an artifact: PDF                | **exists** — venue profile + checker    | 🛑 not allowed into the old form                               |
-| cross-file, filesystem-shaped   | **doesn't exist and isn't expected to** | stays a script — that is the rule's boundary, not a workaround |
-
-**Enforcement:** `paper-lint.mjs` currently has exactly **18** check functions. The "there are 18"
-test fails on the nineteenth and forces an answer to "why isn't this in the new form." Small and
-readable, unlike a 27-entry debt ledger that would get muted within a day.
-
-🔴 **The order is mandatory: freeze AFTER at least half is in the new form.** Right now it's one
-rule out of fifteen. Freezing before the new form becomes more convenient than the old one means
-getting a workaround and a dead rule. That has already happened in this repo three times.
-
----
-
-# 📋 Work queue, cheapest to most expensive
-
-## 1. A marker instead of a blocklist — one line
-
-`paperDirs()` today: `.filter((e) => e.name !== "research" && e.name !== "drafts")`. A hand-written
-list of two names; everything else under `papers/` is automatically counted as a paper.
-
-**Measured 2026-08-26:**
-
-| folder                                              | `PIPELINE-STATUS.md` | counted as a paper today |
-| --------------------------------------------------- | -------------------- | ------------------------ |
-| `agenticdev-2026` `aisec-2026` `compile-rules-2026` | ✅                   | ✅ correctly             |
-| `scored-2026`                                       | —                    | 🔴 **yes, wrongly**      |
-| `extension-shell-ifc`                               | —                    | 🔴 **yes, wrongly**      |
-| `research` `drafts`                                 | —                    | no (on the blocklist)    |
-
-**The marker: a paper folder is one that has a `PIPELINE-STATUS.md`.** This has already been our
-convention since 22.07 ("every paper carries a PIPELINE-STATUS.md"), the code just doesn't use it.
-Self-declaration instead of a list: a new paper enters the checks the day it gets a scorecard.
-
-⚠️ **Do NOT move the three papers into a separate subfolder.** The marker solves the same problem in
-one line, while a move breaks every form of reference — when `.claude/pipeline/` was resettled there
-were 134 literal ones, and that wasn't even the whole count (paths built from segments, relative
-imports, paths inside regexes).
-
-## 2. One build entry point for all papers, not one per paper
-
-> **Done (#59, 2026-09-24).** `paperlint build <paper>` is the entry point, and `paperlint toolchain` installs
-> TeX Live with the packages the venue presets declare; `ensure-toolchain.sh` and
-> `ci-install-texlive.sh` are deleted. What follows is the plan as it was written.
-
-**The hole:** `compile-rules` has `repro/build-submission.sh`, which calls `ensure-toolchain.sh`.
-**`agenticdev` has no entry point at all** — it's built by hand with `pdflatex`, and nothing
-guarantees the fonts except the CI container. That's how the submitted `aisec` went out wrong.
-
-**The fix — generalize the existing script to any paper folder**, instead of giving each one its own
-Makefile. It decides how to build from the files present: `paper.tex` exists → `pdflatex`;
-`paper.md` exists → the jinja2 pipeline. Three Makefiles would drift apart the same way the four
-copies of the page limit and the five copies of the package list already have.
-
-**Don't conflate the layers:**
-
-```
-packages and fonts exist   ← the texlive container  or  ensure-toolchain.sh (apt)
-the compile cycle          ← latexmk (not in the toolchain yet, add it)
-fail if fonts are missing  ← \IfFileExists in paper.tex (done 26.08)
-```
-
-`latexmk` replaces **only the middle layer** — it does not install fonts.
-
-**There's then no need to check "does this folder have a Makefile"** — there's one shared entry
-point. Nothing to check.
-
-## 3. Remove duplicated constants
-
-| fact                 | how many copies | where they diverge                                                                                                                    |
-| -------------------- | --------------- | ------------------------------------------------------------------------------------------------------------------------------------- |
-| page limit           | **4**           | `venues/realm.md` 8 · `build-submission.sh` 8 · `PIPELINE-STATUS` "8/8" · the build prints **9**                                      |
-| the TeX package list | **5**           | `ensure-toolchain.sh` (executable) · `SKILL.md` · `render-paper.harness.mjs` (pins it ✅) · `build-submission.sh` · `PIPELINE-STATUS` |
-
-The canonical source is the venue preset (`presets/<name>.jsonc`, resolved through its `extends`
-chain): its `format` block for the numbers (page size, columns, fonts, font sizes, and `kinds` for
-page limits), its `tex` block (plus `tex-base.jsonc`) for packages — since 2026-09-24;
-`ensure-toolchain.sh` held them until then — and its `rules` block for the checks the venue
-implies. The `.md` card is prose about the venue. Everything else is a pointer.
-
-## 3-ter. Drop the font gate from `check-render.sh` once the rules run in CI
-
-The `pdffonts` block was added to `check-render.sh` on 25.08, and once the judgment moves into rules
-it becomes **a second source of truth about the same fonts**. Removing it shrinks the frozen script
-rather than extending it, so it does not break the freeze.
-
-⚠️ **But not before the rules are actually running in CI.** Today `check-render.sh` is the only
-thing in CI that even looks at `agenticdev-2026` (the `build` job only builds papers that have
-`repro/build-submission.sh`, and that's one out of five). Removing it before the replacement is in
-place means trading a duplicate for a hole.
-
-## 3-bis. 🔴 Remove the venue-profile duplicate by GENERATING it, not guarding it with an assert
-
-The venue numbers live in two places: `<venue>.yaml` (read by the rules) and `<venue>.tex` (read by
-LaTeX itself, which can't do YAML). On 26.08 `texMismatch()` inside `check-geometry.mjs` guarded
-against the drift — **the script was torn out on 26.08, along with it**, because the whole thing was
-rung 5.
-
-**The assert was the wrong answer from the start.** It guards a duplicate instead of there being no
-duplicate. The right form is the same one we arrived at for the PDF facts: **`<venue>.tex` IS
-GENERATED from `<venue>.yaml` in the same step that builds the paper**, lives in `_build/`, and
-never enters git. Then the drift is unrepresentable, and there's nothing to guard — rung 1, not rung 5.
-
-⚠️ The objection "a generated file goes stale silently" **does not apply** here, and the distinction
-matters: what goes stale is a file that's committed and regenerated only occasionally. This one
-regenerates on every build, like the PDF facts do. Silent staleness requires that someone be able to
-read the OLD copy — and there is nowhere to get one.
-
-**Until this is done, NOBODY guards the duplicate.** Recorded deliberately: the cost is a mismatch
-between the preamble guards and the rules; the odds are low — profile edits happen about once per
-venue.
-
-## 4. Tools: what's been checked and decided
-
-|                                            | verdict                                                                                                                                                                                                                                                                                                                                                                                    |
-| ------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
-| **`eslint-plugin-project-structure`**      | **can do** conditional "if A then B" (`enforceExistence`, via a real `fs.existsSync`), and it ships a dummy parser + `files: ["**"]`, so it sees folders with not a single JS file. **But there's no OR semantics** — "Makefile OR justfile" is not expressible. We don't need it after item 2; a candidate for other structural rules                                                     |
-| **`steiger`**                              | ❌ dropped — its own README says _"not extendable with more rules"_, plus it's about FSD in JS                                                                                                                                                                                                                                                                                             |
-| **an ESLint config inside a paper folder** | ❌ **doesn't work in this repository**: there are **233** configs here, **2** of them ours — the other 231 sit inside `compile-rules-2026/repro/` (other people's repos, pulled in). A nested lookup would pick up every one of them, which is exactly why we run with `--no-config-lookup`. The same result comes from a `files: ["<papers-root>/<paper>/**"]` section in the root config |
-| **`banal`**                                | ✅ adopted — fetched by `paperlint toolchain` at a pinned HotCRP commit (GPL; not vendored into paperlint), run on pdf.js-written XML by `paperlint build` and `extract-pdf-facts.mjs` (measurement only; the ESLint rules do the judging)                                                                                                                                                 |
-| **`latexmk`**                              | superseded: `paperlint build` runs its own loop (`src/latex-loop.ts`)                                                                                                                                                                                                                                                                                                                      |