paperlint 4.0.0 → 4.2.0

package/skills/pc-panel-review/SKILL.md

@@ -1,10 +1,10 @@
 ---
 name: pc-panel-review
-description: Use when asking "what would the program committee decide?" / "simulate the reviewers" / "what's this paper's accept probability?" on a drafted paper + artifact. Spawns N independent reviewers with DISTINCT lenses (one actually RUNS the artifact), then synthesizes a PC-chair meta-review into an accept/reject decision, a calibrated probability, and consensus must-fixes. Requires grade-paper-writing's persona stall inventory + tighten-paper's structural verdict as inputs (blocked until they exist — the panel cannot feel reader fatigue on its own). Also has a lightweight single-reviewer VENUE-FIT MODE for a quick CFP-fit spot-check. NOT a single fast defect hunt (paper-adversarial-review), a writing grade (grade-paper-writing), or the full pre-submit gate (harden-paper, which calls this) — this models the PC decision itself.
+description: Use when asking "what would the program committee decide?" / "simulate the reviewers" / "what's this paper's accept probability?" on a drafted paper + artifact. Spawns N independent reviewers with DISTINCT lenses (one actually RUNS the artifact, one is a skimming genre/register reviewer), then synthesizes a PC-chair meta-review into an accept/reject decision reported as min/median/max, a calibrated probability, and consensus must-fixes. Requires grade-paper-writing's persona stall inventory + tighten-paper's structural verdict as inputs (blocked until they exist — the panel cannot feel reader fatigue on its own). Also has a lightweight single-reviewer VENUE-FIT MODE for a quick CFP-fit spot-check. NOT a single fast defect hunt (paper-adversarial-review), a writing grade (grade-paper-writing), or the full pre-submit gate (harden-paper, which calls this) — this models the PC decision itself.
 allowed-tools: [Read, Write, Grep, Glob, Bash, WebSearch, WebFetch, Agent, Skill]
 ---
 
-<!-- vigiles:sha256:b9feb099347291ff compiled from skills/pc-panel-review/SKILL.md.spec.ts -->
+<!-- vigiles:sha256:7e5fb04d1cfa0548 compiled from skills/pc-panel-review/SKILL.md.spec.ts -->
 
 # pc-panel-review — model the whole PC, not one reviewer
 
@@ -51,7 +51,15 @@ prose-only read can.
    panel's only legitimate source for the readability/structure signal.
    *(Scorecard: this gate is row **`panel`**, and it declares **`structure`** (tighten-paper) + **`writing`**
    (grade-paper-writing) as its required inputs.)*
-3. **Pick 3 (or 4) reviewer lenses** that a real PC for THIS venue would field. Menu — choose by fit:
+3. **Pick 3 (or 4) reviewer lenses** that a real PC for THIS venue would field. One is fixed; the rest
+   come from the menu, by fit:
+   - **Genre/register reviewer — MANDATORY on every panel, at every venue.** Confidence 3 (knows the
+     area, not this topic), reads ONLY the abstract, the introduction and the conclusion, and judges
+     whether this reads as a paper at this venue: whose result each sentence states, whether the
+     relations between claims are signposted or left to infer, and whether anything is an aphorism,
+     a fragment or a colloquial turn. This is the reviewer who writes one short review, scores low,
+     and calls the paper a blog post — the one reviewer a panel of careful full readers never
+     predicts. Give it the three sections as text and the venue's name, nothing else.
    - **Methods/stats skeptic** — sample size, construct validity, multiple comparisons, overclaims,
      whether the CIs support the claims.
    - **Domain practitioner** — relevance to the CFP, novelty vs the nearest prior work, does the
@@ -67,7 +75,13 @@ prose-only read can.
 4. **Spawn them in PARALLEL, each on a separate model instance** (e.g. Fable via subagent), each blind
    to the others — independence is the whole point. Give each: the paper (`.tex`/PDF), the artifact
    path, the quoted rubric, its lens, **and the two step-2 artifacts (persona stall inventory +
-   tighten-paper structural verdict)**. Force a filled review form (below).
+   tighten-paper structural verdict)** — the genre/register reviewer gets only its three sections.
+   Force a filled review form (below).
+
+   🔴 **No reviewer is shown a previous round** — not its scores, not its reviews, not "the panel
+   gave 3.25 last time", not a list of what changed since. A reviewer who is told the last score
+   anchors on it (*"up from the 3.25 the earlier rounds reported"*), and the rounds then converge on
+   each other instead of on the paper. Comparing rounds is the chair's job, after the reviews land.
 5. **Synthesize the meta-review yourself** (PC chair) once all land — see output.
 
 ## Each reviewer returns a review form
@@ -90,10 +104,19 @@ prose-only read can.
 - One-line meta: does this belong at this venue?
 
 ## The meta-review (PC-chair synthesis) — the actual deliverable
-- **Score table:** each reviewer × each criterion + overall + confidence, at a glance.
-- **Decision:** the rec the PC discussion converges on (weight by confidence; a lone low-confidence
-  outlier doesn't sink two confident accepts), + a **calibrated accept probability** for this venue
-  and edition.
+- **Score table:** each reviewer × each criterion + overall + confidence, at a glance, and under it
+  the overall score's **min / median / max** across reviewers — never a mean alone, which hides the
+  one low review that decides a workshop.
+- **Decision:** for a **workshop**, the decision follows the **MIN**: a workshop PC with three
+  reviews and a short discussion does not overrule a reject, so one Reject or Strong Reject sinks the
+  paper however confident the others are. For a conference, the rec the PC discussion converges on,
+  weighted by confidence, with the min named beside it. Either way, a **calibrated accept probability**
+  for this venue and edition.
+- 🔴 **One reviewer is not a panel.** A run with fewer than three reviews and a meta-review — a
+  Venue-fit spot-check, a single re-review, a panel whose other seats did not return — is recorded as
+  what it is (`venue-fit`, `single review`), never in the scorecard's `panel` row. A single
+  review recorded as a panel verdict is how a paper reaches submission on an 88% that three real
+  reviewers turned into a reject.
 - **Consensus must-fixes:** issues **≥2 reviewers independently raised** — these are the real ones;
   apply before submitting.
 - **Single-reviewer flags:** noted, triaged (fix if cheap, else defer to camera-ready).
@@ -199,9 +222,11 @@ reject-risk without rubber-stamping. For the real decision, escalate to the full
 ## The reproduce loop (why this beats a prose review)
 This is "double-blind review as a real venue does it, WITH reproduction." The artifact-runner reviewer
 is the difference-maker on both runs so far: it doesn't judge the paper's numbers, it re-derives them.
-Run the panel, apply the consensus fixes, then **re-run the panel** (same lenses, told what changed) to
-confirm the fixes landed and nothing regressed — the artifact-runner re-executes every harness each
-round. Track the accept-probability progression in the review ledger. Stop when the panel converges to
+Run the panel, apply the consensus fixes, then **re-run the panel** — same lenses, FRESH reviewers,
+told nothing about the previous round — to confirm the fixes landed and nothing regressed; the
+artifact-runner re-executes every harness each round. The chair, not the reviewers, sets the new
+reviews beside the old ones and checks that each earlier must-fix is gone. Track the min / median / max
+progression in the review ledger. Stop when the panel converges to
 accept and the artifact reproduces clean; don't loop past one confirming round.
 
 ## Provenance (two papers, battle-tested)

package/skills/pc-panel-review/SKILL.md.spec.ts

@@ -9,7 +9,7 @@ import { experimental_skill } from "vigiles/spec";
 export default experimental_skill({
   name: "pc-panel-review",
   description:
-    'Use when asking "what would the program committee decide?" / "simulate the reviewers" / "what\'s this paper\'s accept probability?" on a drafted paper + artifact. Spawns N independent reviewers with DISTINCT lenses (one actually RUNS the artifact), then synthesizes a PC-chair meta-review into an accept/reject decision, a calibrated probability, and consensus must-fixes. Requires grade-paper-writing\'s persona stall inventory + tighten-paper\'s structural verdict as inputs (blocked until they exist — the panel cannot feel reader fatigue on its own). Also has a lightweight single-reviewer VENUE-FIT MODE for a quick CFP-fit spot-check. NOT a single fast defect hunt (paper-adversarial-review), a writing grade (grade-paper-writing), or the full pre-submit gate (harden-paper, which calls this) — this models the PC decision itself.',
+    'Use when asking "what would the program committee decide?" / "simulate the reviewers" / "what\'s this paper\'s accept probability?" on a drafted paper + artifact. Spawns N independent reviewers with DISTINCT lenses (one actually RUNS the artifact, one is a skimming genre/register reviewer), then synthesizes a PC-chair meta-review into an accept/reject decision reported as min/median/max, a calibrated probability, and consensus must-fixes. Requires grade-paper-writing\'s persona stall inventory + tighten-paper\'s structural verdict as inputs (blocked until they exist — the panel cannot feel reader fatigue on its own). Also has a lightweight single-reviewer VENUE-FIT MODE for a quick CFP-fit spot-check. NOT a single fast defect hunt (paper-adversarial-review), a writing grade (grade-paper-writing), or the full pre-submit gate (harden-paper, which calls this) — this models the PC decision itself.',
   tools: [
     "Read",
     "Write",
@@ -67,7 +67,15 @@ prose-only read can.
    panel's only legitimate source for the readability/structure signal.
    *(Scorecard: this gate is row **\`panel\`**, and it declares **\`structure\`** (tighten-paper) + **\`writing\`**
    (grade-paper-writing) as its required inputs.)*
-3. **Pick 3 (or 4) reviewer lenses** that a real PC for THIS venue would field. Menu — choose by fit:
+3. **Pick 3 (or 4) reviewer lenses** that a real PC for THIS venue would field. One is fixed; the rest
+   come from the menu, by fit:
+   - **Genre/register reviewer — MANDATORY on every panel, at every venue.** Confidence 3 (knows the
+     area, not this topic), reads ONLY the abstract, the introduction and the conclusion, and judges
+     whether this reads as a paper at this venue: whose result each sentence states, whether the
+     relations between claims are signposted or left to infer, and whether anything is an aphorism,
+     a fragment or a colloquial turn. This is the reviewer who writes one short review, scores low,
+     and calls the paper a blog post — the one reviewer a panel of careful full readers never
+     predicts. Give it the three sections as text and the venue's name, nothing else.
    - **Methods/stats skeptic** — sample size, construct validity, multiple comparisons, overclaims,
      whether the CIs support the claims.
    - **Domain practitioner** — relevance to the CFP, novelty vs the nearest prior work, does the
@@ -83,7 +91,13 @@ prose-only read can.
 4. **Spawn them in PARALLEL, each on a separate model instance** (e.g. Fable via subagent), each blind
    to the others — independence is the whole point. Give each: the paper (\`.tex\`/PDF), the artifact
    path, the quoted rubric, its lens, **and the two step-2 artifacts (persona stall inventory +
-   tighten-paper structural verdict)**. Force a filled review form (below).
+   tighten-paper structural verdict)** — the genre/register reviewer gets only its three sections.
+   Force a filled review form (below).
+
+   🔴 **No reviewer is shown a previous round** — not its scores, not its reviews, not "the panel
+   gave 3.25 last time", not a list of what changed since. A reviewer who is told the last score
+   anchors on it (*"up from the 3.25 the earlier rounds reported"*), and the rounds then converge on
+   each other instead of on the paper. Comparing rounds is the chair's job, after the reviews land.
 5. **Synthesize the meta-review yourself** (PC chair) once all land — see output.
 
 ## Each reviewer returns a review form
@@ -106,10 +120,19 @@ prose-only read can.
 - One-line meta: does this belong at this venue?
 
 ## The meta-review (PC-chair synthesis) — the actual deliverable
-- **Score table:** each reviewer × each criterion + overall + confidence, at a glance.
-- **Decision:** the rec the PC discussion converges on (weight by confidence; a lone low-confidence
-  outlier doesn't sink two confident accepts), + a **calibrated accept probability** for this venue
-  and edition.
+- **Score table:** each reviewer × each criterion + overall + confidence, at a glance, and under it
+  the overall score's **min / median / max** across reviewers — never a mean alone, which hides the
+  one low review that decides a workshop.
+- **Decision:** for a **workshop**, the decision follows the **MIN**: a workshop PC with three
+  reviews and a short discussion does not overrule a reject, so one Reject or Strong Reject sinks the
+  paper however confident the others are. For a conference, the rec the PC discussion converges on,
+  weighted by confidence, with the min named beside it. Either way, a **calibrated accept probability**
+  for this venue and edition.
+- 🔴 **One reviewer is not a panel.** A run with fewer than three reviews and a meta-review — a
+  Venue-fit spot-check, a single re-review, a panel whose other seats did not return — is recorded as
+  what it is (\`venue-fit\`, \`single review\`), never in the scorecard's \`panel\` row. A single
+  review recorded as a panel verdict is how a paper reaches submission on an 88% that three real
+  reviewers turned into a reject.
 - **Consensus must-fixes:** issues **≥2 reviewers independently raised** — these are the real ones;
   apply before submitting.
 - **Single-reviewer flags:** noted, triaged (fix if cheap, else defer to camera-ready).
@@ -215,9 +238,11 @@ reject-risk without rubber-stamping. For the real decision, escalate to the full
 ## The reproduce loop (why this beats a prose review)
 This is "double-blind review as a real venue does it, WITH reproduction." The artifact-runner reviewer
 is the difference-maker on both runs so far: it doesn't judge the paper's numbers, it re-derives them.
-Run the panel, apply the consensus fixes, then **re-run the panel** (same lenses, told what changed) to
-confirm the fixes landed and nothing regressed — the artifact-runner re-executes every harness each
-round. Track the accept-probability progression in the review ledger. Stop when the panel converges to
+Run the panel, apply the consensus fixes, then **re-run the panel** — same lenses, FRESH reviewers,
+told nothing about the previous round — to confirm the fixes landed and nothing regressed; the
+artifact-runner re-executes every harness each round. The chair, not the reviewers, sets the new
+reviews beside the old ones and checks that each earlier must-fix is gone. Track the min / median / max
+progression in the review ledger. Stop when the panel converges to
 accept and the artifact reproduces clean; don't loop past one confirming round.
 
 ## Provenance (two papers, battle-tested)

package/skills/render-paper/SKILL.md

@@ -4,7 +4,7 @@ description: Compile a LaTeX paper (ACM/IEEE/arXiv) to PDF and render its pages
 allowed-tools: [Read, Write, Edit, Grep, Glob, Bash, SendUserFile]
 ---
 
-<!-- vigiles:sha256:a1ec1d392f144c80 compiled from skills/render-paper/SKILL.md.spec.ts -->
+<!-- vigiles:sha256:5e2ee8850c098375 compiled from skills/render-paper/SKILL.md.spec.ts -->
 
 # render-paper — .tex → PDF → readable page PNGs
 
@@ -34,7 +34,7 @@ That list was replaced on 2026-08-03; nothing about the toolchain below changed.
   by `kpsewhich` against the files each profile names. The last line it prints is `bin: <dir>` —
   put that directory first on PATH before calling `pdflatex` or `check-render.sh` by hand.
 
-  🔴 The package list lives in the venue profiles (`submit-paper/references/venues/*.jsonc`, the
+  🔴 The package list lives in the venue profiles (`presets/*.jsonc` in the package, the
   `tex` block; `tex-base.jsonc` for every paper) and nowhere else. A new venue that needs a new
   package gets it there — never in a script, never here. A TeX Live without `libertine` builds an
   acmart paper GREEN in Computer Modern; that is why the files are checked, not the exit code.

package/skills/render-paper/SKILL.md.spec.ts

@@ -40,7 +40,7 @@ That list was replaced on 2026-08-03; nothing about the toolchain below changed.
   by \`kpsewhich\` against the files each profile names. The last line it prints is \`bin: <dir>\` —
   put that directory first on PATH before calling \`pdflatex\` or \`check-render.sh\` by hand.
 
-  🔴 The package list lives in the venue profiles (\`submit-paper/references/venues/*.jsonc\`, the
+  🔴 The package list lives in the venue profiles (\`presets/*.jsonc\` in the package, the
   \`tex\` block; \`tex-base.jsonc\` for every paper) and nowhere else. A new venue that needs a new
   package gets it there — never in a script, never here. A TeX Live without \`libertine\` builds an
   acmart paper GREEN in Computer Modern; that is why the files are checked, not the exit code.

package/skills/render-paper/check-render.sh

@@ -25,75 +25,28 @@ set -uo pipefail
 # check — that is, the defect the guard was set up for comes back together with a green report.
 # A broken `import` in JS fails loudly; here we have to fail on our own.
 #
-# A ladder of THREE candidates, because the directory MOVED (12.09): `venues/` is now carried by the
-# `paperlint` package, and at the consumer's old location there is a symlink into it.
-# The order "the declared package → a sibling inside the package → the consumer's directory" is
-# deliberate: the package must win silently. It only gets loud when there is NOT A SINGLE ONE.
-# 🔴 THE SCRIPT'S OWN DIRECTORY — TAKEN ABSOLUTE AND BEFORE ANY `cd` (12.09.2026, evening).
-# Found by a RED HARNESS, `gates.harness.mjs`, not by proofreading, and it is my own regression from
-# the same day: below, the script does `cd "$DIR"` into the paper directory, and the venues resolve
-# stood AFTER that `cd`. So `git rev-parse --show-toplevel` was already running in the fixture's
-# temporary directory, the root was not found, and the build honestly stopped — on a test that had
-# been passing until then.
+# TWO candidates, in this order: the package by its NAME, from the current directory (a consumer's
+# installed paperlint wins), then the package THIS script ships in. Loud only when neither holds it.
 #
-# ⚠️ The previous version was wrong THE SAME WAY, but silently: it did not check the result and built
-# the PDF without the reference guard. That is, going red is not a new breakage but a manifestation
-# of an old one; still, a test that was green yesterday and is red today is mine to fix.
-#
-# The neighbouring `build.sh` already carries this same lesson in a comment ("The script directory is
-# taken ONCE and ABSOLUTE, before cd") — CI failed on it there on 30.08. A second instance of the same
-# class in a file right next door.
+# 🔴 THE SCRIPT'S OWN DIRECTORY IS TAKEN ABSOLUTE AND BEFORE ANY `cd`: below, the script `cd`s into
+# the paper directory, and a lookup after that would start from the paper, not from here.
 SELF_DIR=$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)
 
 resolve_venues() {
-  local root pkg
-  # Three sources for the root, from the most explicit to the most reliable. The last one is the
-  # SCRIPT's location (`.claude/skills/render-paper/` ⇒ three levels up), and it is the only one that
-  # depends neither on the current directory nor on whether the caller sits inside a git tree.
-  # ⚠️ THE PARENTHESES ARE MANDATORY, and the harness caught this a minute after I wrote the line
-  # without them. In bash `A || B && C` parses as `(A || B) && C`, so on a SUCCESSFUL `git rev-parse`
-  # the `pwd` ran as well, and the substitution returned TWO lines — the repository root plus the
-  # current directory. By hand from a directory outside git this did not reproduce: there git failed,
-  # and only one branch ran.
-  # ⚠️ THE LAST RUNG SEARCHES, IT DOES NOT COUNT LEVELS. `cd "$SELF_DIR/../../.."` used to stand here
-  # — correct for `.claude/skills/render-paper/` (three levels to the root) and silently wrong after
-  # the skill moved into the package, where it lies in `skills/render-paper/` (two): the climb
-  # overshot ABOVE the repository, and rung 2 looked for venues in someone else's directory. Climbing
-  # up to the directory that holds `.claude` does not depend on the depth.
-  local probe="$SELF_DIR" fallback=""
-  while [ "$probe" != "/" ]; do
-    if [ -d "$probe/.claude" ]; then fallback="$probe"; break; fi
-    probe=$(dirname "$probe")
-  done
-  root="${CLAUDE_PROJECT_DIR:-$(git rev-parse --show-toplevel 2>/dev/null || printf '%s' "$fallback")}"
-
-  # 1) the package, if it already carries venues/ (after step 5)
-  pkg=$(node -e 'try{process.stdout.write(require.resolve("paperlint/venues/paper-guards.tex"))}catch{}' 2>/dev/null || true)
+  local pkg own
+  # 1) the package by its name. `node -e` resolves from the current directory.
+  pkg=$(node -e 'try{process.stdout.write(require.resolve("paperlint/presets/tex/paper-guards.tex"))}catch{}' 2>/dev/null || true)
   if [ -n "$pkg" ] && [ -f "$pkg" ]; then
     dirname "$pkg"
     return 0
   fi
 
-  # 2) THE SIBLING SKILL, from THIS script's directory. Added 12.09 together with the move of
-  # `submit-paper` into the package, and it is the only one of the three that works in BOTH worlds at
-  # once:
-  #   • package checkout   `skills/render-paper/..` → `skills/submit-paper/references/venues`
-  #   • consumer           `.claude/skills/render-paper/..` → `.claude/skills/submit-paper/...`
-  #     (both directories are symlinks into the package, `-f` goes through them, and `cd`+`pwd`
-  #     leaves the logical path)
-  # Rung 1 depends on cwd (`node -e` resolves from the current directory), rung 3 depends on the
-  # consumer's root having been found at all. This one depends on neither — exactly the anchor whose
-  # absence had `gates.harness.mjs` red on 12.09.
-  local sibling="$SELF_DIR/../submit-paper/references/venues"
-  if [ -f "$sibling/paper-guards.tex" ]; then
-    (cd "$sibling" && pwd)
-    return 0
-  fi
-
-  # 3) the skill directory in the consumer — for a consumer that mounted the skills differently
-  local local_dir="$root/.claude/skills/submit-paper/references/venues"
-  if [ -f "$local_dir/paper-guards.tex" ]; then
-    printf '%s\n' "$local_dir"
+  # 2) the package this script is part of: `skills/render-paper/` → `presets/tex/`. `cd -P` follows
+  # the symlinks a consumer's `.claude/skills/render-paper` is made of, so `../..` is the PACKAGE's
+  # root, not the consumer's `.claude/`. Depends neither on the cwd nor on a git tree.
+  own="$(cd -P "$SELF_DIR" && pwd)/../../presets/tex"
+  if [ -f "$own/paper-guards.tex" ]; then
+    (cd "$own" && pwd)
     return 0
   fi
 
@@ -107,8 +60,8 @@ if [ "${1:-}" = "--print-venues" ]; then
     printf '%s\n' "$VD"
     exit 0
   fi
-  echo "✗ paper-guards.tex not found: not under the package name paperlint, not beside the script" >&2
-  echo "  (skills/submit-paper/references/venues), not at the consumer (.claude/skills/submit-paper/references/venues)." >&2
+  echo "✗ paper-guards.tex not found: not under the package name paperlint (paperlint/presets/tex)," >&2
+  echo "  not in the package this script ships in (../../presets/tex from it)." >&2
   echo "  Building the paper without it is not allowed: \\input{paper-guards} will not fail, it will vanish silently," >&2
   echo "  and the PDF will be built WITHOUT the check for dangling \\ref and \\cite." >&2
   exit 2
@@ -135,7 +88,7 @@ if [ "${NO_COMPILE:-0}" != "1" ] && [ -f "$BASE.tex" ]; then
   # Two passes: overfull boxes surface on every pass; a second pass settles refs
   # so Reference/Citation-undefined warnings are accurate.
   # 🔴 TEXINPUTS is mandatory, and that was found by a crash on 26.08. Papers pull in their venue's
-  # numbers via `\input{<venue>}`, and the file lies in `.claude/skills/submit-paper/references/venues/`.
+  # numbers via `\input{<venue>}`, and the file lies in the package's `presets/tex/`.
   # Without that path the build fails with `File \`agenticdev.tex' not found` — and the `|| true`
   # below swallows it, after which the script reads a log of UNKNOWN origin and reports on it. That
   # is, compiling this paper here silently did not work, and it only became visible once a pass

package/skills/render-paper/render-paper.harness.mjs

@@ -68,34 +68,11 @@ await checkSkill("render-paper");
   const run = (args, opts = {}) =>
     spawnSync("bash", [CHECK, ...args], { encoding: "utf8", ...opts });
 
-  // First half — on TODAY's tree, resolution must succeed IF THERE IS ANYTHING TO
-  // resolve.
-  //
-  // 🔴 THERE USED TO BE AN UNCONDITIONAL `assert.equal(status, 0)` HERE, AND THE MOVE BROKE IT —
-  // for a real reason, not by accident. It relied on the SURROUNDING tree: while the skill lived
-  // at the consumer, the second rung of the ladder won
-  // (`.claude/skills/submit-paper/references/venues/`). The package checkout has neither that nor
-  // the first rung — `venues/` arrives together with `submit-paper`, which hasn't moved yet. So
-  // the assertion wasn't checking resolution, it was checking for the presence of an unrelated
-  // directory nearby.
-  //
-  // The form now is: resolution must succeed IF AND ONLY IF at least one candidate exists on
-  // disk. This is still a real assertion — it fails if a candidate exists and the script didn't
-  // find it — but it stops requiring the existence of something this repository doesn't have yet.
-  // The absence of both candidates gets PRINTED, not swallowed: "the check found nothing to
-  // check" and "the check passed" have to look different.
+  // First half — on this tree, resolution must succeed IF AND ONLY IF a candidate exists on disk:
+  // the package's own `presets/tex/`. The absence of it is PRINTED, not swallowed: "the check found
+  // nothing to check" and "the check passed" have to look different.
   const candidates = [
-    // rung 1 — the package
-    join(HERE, "..", "..", "venues", "paper-guards.tex"),
-    // rung 2 — the skill directory at the consumer
-    join(
-      HERE,
-      "..",
-      "submit-paper",
-      "references",
-      "venues",
-      "paper-guards.tex",
-    ),
+    join(HERE, "..", "..", "presets", "tex", "paper-guards.tex"),
   ].filter((c) => existsSync(c));
 
   const ok = run(["--print-venues"]);
@@ -108,9 +85,8 @@ await checkSkill("render-paper");
         `silent success this whole block was written to catch.`,
     );
     console.log(
-      `  --print-venues: NOT VERIFIED on the live tree — paper-guards.tex is absent from both ` +
-        `the package (venues/) and next to it (submit-paper/references/venues/). It arrives with ` +
-        `\`submit-paper\`; until then only the fixtures below hold this half.`,
+      `  --print-venues: NOT VERIFIED on the live tree — paper-guards.tex is absent from the ` +
+        `package's presets/tex/; only the fixtures below hold this half.`,
     );
   } else {
     assert.equal(
@@ -159,18 +135,9 @@ await checkSkill("render-paper");
   // When the file is NOWHERE, the script must exit with code 2 and say why. A silent success here
   // is exactly the defect the whole ladder was written for.
   //
-  // 🔴 RUN AGAINST A COPY OF THE SCRIPT, AND THIS IS NOT A TEST WORKAROUND, IT'S A CONSEQUENCE OF
-  // THE MOVE (09-12, wave ③). Before `submit-paper` moved, "nowhere" was reproduced by an empty
-  // consumer root: both former rungs looked outward (the package name from cwd · `.claude/` under
-  // the root), and an empty directory found nothing at either. Now the script has a rung anchored
-  // to ITS OWN directory (`$SELF_DIR/../submit-paper/references/venues`) — added for
-  // cwd-independence — so as long as the script sits inside the package, venues next to it exist
-  // ALWAYS, and "nowhere" can no longer be reached through the environment. The only honest way to
-  // demonstrate that state is to move the script itself out of the package. The property being
-  // checked is otherwise identical, word for word: nothing resolved ⇒ exit 2 + the name of the
-  // missing file. ⚠️ What this form no longer checks: that an empty consumer root won't make the
-  // script lie — and it shouldn't, because such a root is now legitimately shadowed by the copy in
-  // the package.
+  // 🔴 RUN AGAINST A COPY OF THE SCRIPT. Inside the package its second rung (the package it ships
+  // in) always holds the file, so "nowhere" is reachable only by moving the script out of the
+  // package: nothing resolved ⇒ exit 2 + the name of the missing file.
   const empty = realpathSync(mkdtempSync(join(tmpdir(), "venues-none-")));
   const lonely = join(empty, "check-render.sh");
   copyFileSync(CHECK, lonely);
@@ -190,30 +157,21 @@ await checkSkill("render-paper");
     "the failure message must name the missing file",
   );
 
-  // Third half — the ORDER of the rungs. Not cosmetic: on the day of the move, both rungs will be
-  // true at once, and the package has to win, otherwise the consumer silently keeps building with
-  // their old copy. Plant BOTH candidates and see which one is chosen.
+  // Third half — the ORDER of the rungs. The package the consumer installed, found by its name,
+  // must win over the copy this script ships in; otherwise a consumer silently builds with a stale
+  // copy. Plant a package by name beside the real script and see which one is chosen.
   const both = realpathSync(mkdtempSync(join(tmpdir(), "venues-both-")));
-  const pkgVenues = join(both, "node_modules", "paperlint", "venues");
-  const skillVenues = join(
-    both,
-    ".claude",
-    "skills",
-    "submit-paper",
-    "references",
-    "venues",
-  );
-  for (const d of [pkgVenues, skillVenues]) mkdirSync(d, { recursive: true });
+  const pkgVenues = join(both, "node_modules", "paperlint", "presets", "tex");
+  mkdirSync(pkgVenues, { recursive: true });
   writeFileSync(
     join(both, "node_modules", "paperlint", "package.json"),
     JSON.stringify({
       name: "paperlint",
       version: "0.0.0",
-      exports: { "./venues/*": "./venues/*" },
+      exports: { "./*": "./*" },
     }),
   );
   writeFileSync(join(pkgVenues, "paper-guards.tex"), "% from package\n");
-  writeFileSync(join(skillVenues, "paper-guards.tex"), "% from skill dir\n");
   const race = run(["--print-venues"], {
     cwd: both,
     env: { ...process.env, CLAUDE_PROJECT_DIR: both },
@@ -226,7 +184,7 @@ await checkSkill("render-paper");
   assert.equal(
     race.stdout.trim(),
     pkgVenues,
-    "with both candidates present, the PACKAGE wins, not the skill directory",
+    "with both candidates present, the INSTALLED package wins, not the copy the script ships in",
   );
 }
 
@@ -276,7 +234,7 @@ await checkSkill("render-paper");
     );
     assert.match(
       build,
-      /require\.resolve\("paperlint\/venues\/paper-guards\.tex"\)/,
+      /require\.resolve\("paperlint\/presets\/tex\/paper-guards\.tex"\)/,
       `${where}: the ladder's first rung (the package) must remain`,
     );
     assert.match(

package/src/adapters/banal/index.test.ts

@@ -2,7 +2,7 @@
  * `adapters/banal/index.ts` — banal as the two ports the app sees. What `run.test.ts` checks in
  * banal's own terms is checked here as the domain receives it: a `Geometry` whose provenance names
  * banal and the rule that found it, a reason that is one line of text, a `Ready` that says what was
- * verified. In-memory ports throughout; the real banal is `test/e2e/banal.ts`.
+ * verified. In-memory ports throughout; the real banal is `test/e2e/tex/banal.e2e.ts`.
  */
 import assert from "node:assert/strict";
 import { test } from "vitest";

package/src/adapters/banal/invocation.ts

@@ -7,7 +7,7 @@
  * (`Error: Failed to run pdftohtml`, measured); told a version below 0.85 it applies a larger
  * correction and every font size moves. So `$PDFTOHTML` (banal's own override, line 166) points at a
  * stub that answers `-v` with `XML_DIALECT.version` — the pdftohtml whose XML `xml.ts` writes — and
- * refuses anything else. `test/e2e/banal.ts` shows both failures with the real banal.
+ * refuses anything else. `test/e2e/tex/banal.e2e.ts` shows both failures with the real banal.
  */
 import type { Opaque } from "ts-essentials";
 import type { AbsolutePath } from "../../domain/paths.ts";

package/src/adapters/banal/output.test.ts

@@ -1,6 +1,6 @@
 /**
  * `output.ts` — every way a banal run can end, as a table of `ProcessExit` → `Result`, and the
- * measurement → geometry projection. Pure: no perl, no disk. The real banal is `test/e2e/banal.ts`.
+ * measurement → geometry projection. Pure: no perl, no disk. The real banal is `test/e2e/tex/banal.e2e.ts`.
  */
 import assert from "node:assert/strict";
 import { test } from "vitest";

package/src/adapters/banal/run.test.ts

@@ -1,7 +1,7 @@
 /**
  * `adapters/banal/run.ts` — running and installing banal, on in-memory ports: which `Command` it builds, what it stages,
  * what it downloads and writes, and what it concludes. No perl, no disk, no network: the adapters
- * record, and each assertion reads the record. Real perl and the real banal are `test/e2e/banal.ts`;
+ * record, and each assertion reads the record. Real perl and the real banal are `test/e2e/tex/banal.e2e.ts`;
  * the adapters' own behaviour is `src/adapters/node/*.test.ts`.
  */
 import assert from "node:assert/strict";

package/src/adapters/banal/xml.test.ts

@@ -4,7 +4,7 @@
  *
  * Every page here is written by hand. The real-PDF half — pdf.js boxes of a committed fixture,
  * written as XML, measured by the real banal and compared with banal on real pdftohtml — is
- * `test/e2e/banal.ts`, which runs where `paperlint toolchain` has installed banal.
+ * `test/e2e/tex/banal.e2e.ts`, which runs where `paperlint toolchain` has installed banal.
  */
 import assert from "node:assert/strict";
 import { test } from "vitest";