paperlint 5.0.0 → 6.1.0

package/fixtures/build-e2e/anon-pdfauthor/expect.json

@@ -0,0 +1,19 @@
+{
+  "$comment": "anonymity/identity on a real IEEEtran build: D: the name only in the PDF metadata, set by hyperref; no page shows it. Exactly one finding, naming where.",
+  "build": {
+    "outcome": "pdf"
+  },
+  "lint": {
+    "rules": {
+      "anonymity/identity": [
+        {
+          "severity": 2,
+          "says": [
+            "PDF metadata: Author",
+            "«Ada Example»"
+          ]
+        }
+      ]
+    }
+  }
+}

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

@@ -0,0 +1,31 @@
+% D: the name only in the PDF metadata, set by hyperref; no page shows it.
+% The declared identity is invented: "Ada Example", handle "adaexample" (paperlint.json).
+\documentclass[conference,compsoc]{IEEEtran}
+\usepackage{hyperref}
+\hypersetup{pdfauthor={Ada Example}}
+\begin{filecontents*}{refs.bib}
+@article{knuth84,
+  title   = {Literate Programming},
+  author  = {Knuth, Donald E.},
+  journal = {The Computer Journal},
+  volume  = {27},
+  number  = {2},
+  pages   = {97--111},
+  year    = {1984},
+  doi     = {10.1093/comjnl/27.2.97}
+}
+\end{filecontents*}
+\begin{document}
+\title{A Fixture for Double-Blind Review}
+\author{\IEEEauthorblockN{Anonymous Authors}}
+\maketitle
+\section{Introduction}
+This paper builds on earlier work~\cite{knuth84}.
+
+\section{Conclusion}
+The conclusion restates that the fixture builds.
+\bibliographystyle{IEEEtran}
+\bibliography{refs}
+\section*{LLM Usage Statement}
+No LLMs were used in writing this fixture.
+\end{document}

package/fixtures/build-e2e/anon-pdfauthor/paperlint.json

@@ -0,0 +1 @@
+{"extends": "paperlint:aidc", "kind": "short", "identity": ["Ada Example", "adaexample", "ada@example.org", "Example University", "exampletool"]}

package/fixtures/build-e2e/unbalanced/expect.json

@@ -18,7 +18,7 @@
     }
   },
   "facts": {
-    "schema": 2,
+    "schema": 3,
     "lastPageColumnsPt": [
       621.5,
       264.8
@@ -45,7 +45,7 @@
       "pdf/profile": [],
       "pdf/fonts": [],
       "pdf/geometry": [],
-      "pdf/limits": [],
+      "format/page-limit": [],
       "pdf/body-size": [],
       "pdf/measured": [
         {

package/lib/paper-config.ts

@@ -136,8 +136,10 @@ export const SETTINGS_KEYS: Readonly<Record<string, string>> = Object.freeze({
     "rule overrides: { id: severity } for every paper in scope, or ESLint blocks (files, ignores, rules) — paperlint lint",
   extends:
     "the venue preset: paperlint:<name> or ./path.jsonc — the pdf/ venue rules, paperlint build",
-  kind: "the kind of paper, whose page limit applies — pdf/limits",
+  kind: "the kind of paper, whose page limit applies — format/page-limit",
   pdf: "where the built PDF is, relative to the paper, when it is not paper.pdf — the facts",
+  identity:
+    "what identifies the authors — names, handles, emails, affiliations, own project names; a blind venue's PDF must not say any of them — anonymity/identity. The root's list and a paper's are joined",
   scripts: "how skill prose names the scripts — consumer.mjs",
   timezone: "the zone deadline anchors are written in — consumer.mjs",
   contactEmail: "the polite-pool address for citation lookups — consumer.mjs",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "paperlint",
-  "version": "5.0.0",
+  "version": "6.1.0",
   "description": "A linter for scientific papers written in LaTeX: catches mistakes before you submit to a conference or journal.",
   "type": "module",
   "license": "MIT",

package/presets/acm-sigconf.jsonc

@@ -74,7 +74,9 @@
     // «font size is 9 pt for all body text»
     "body_pt": 9,
     // banal measures the MODE of the rendered text, not the declared size (measured 08-26: 9.30 at
-    // acmart's declared 9). Hence the tolerance — and hence why body size stays a nudge, not a gate.
+    // acmart's declared 9). Hence the tolerance: banal reads 0.3 pt high on every template measured
+    // (fixtures/accepted-papers/body-size.json), and 0.5 absorbs that while a paper one size off is
+    // still reported.
     "body_pt_tol": 0.5,
 
     // «font size of the references can be from 7 pt to 9 pt»

package/presets/aidc.jsonc

@@ -22,6 +22,10 @@
 {
   "extends": "paperlint:ieee-conference",
 
+  // «Submissions must be properly anonymized for double-blind review.» (the call above), and
+  // `"blind":true` in the portal's data. anonymity/identity then requires the paper's `identity`.
+  "blind": true,
+
   // What the venue is called in a paper's text (tex/venue-leftover): the workshop, and the
   // conference it is held at.
   "aliases": ["AIDC", "ACSAC"],
@@ -39,35 +43,80 @@
   // for) is not a section with this title and does not count.
   "required_sections": [{ "title": "LLM Usage Statement", "position": "last" }],
 
+  // THE REGISTER OF THE BODY IS JUDGED AGAINST ACSAC, AIDC'S PARENT CONFERENCE: AIDC 2026 is the
+  // workshop's first edition, so it has no accepted papers yet, and it uses ACSAC's template. The
+  // anchors are the five accepted ACSAC papers in fixtures/accepted-papers/, measured by paperlint's
+  // own counters on their LaTeX source (tex/contrast-frames, tex/claim-emphasis, tex/relation-markers).
+  // src/register-bands.test.ts re-measures every anchor that has a fixture and requires these exact
+  // numbers, so a change to a counter changes this list in the same commit. The bands they set, per
+  // 10,000 words (the anchors' range, widened by two Poisson standard deviations at each end):
+  //   contrast frames   anchors 1.3–5.5     band 0–10.4
+  //   claims in bold    anchors 0.0–10.4    band 0–17.8
+  //   relation markers  anchors 39.4–112.8  band 25.4–135.0
+  "register": {
+    "anchors": [
+      {
+        "paper": "barovox-acsac24",
+        "words": 7670,
+        "contrast_frames": 1,
+        "claim_emphasis": 8,
+        "relation_markers": 31,
+      },
+      {
+        "paper": "leaking-queries-acsac25",
+        "words": 9133,
+        "contrast_frames": 5,
+        "claim_emphasis": 0,
+        "relation_markers": 103,
+      },
+      {
+        "paper": "llm-splained-acsac25",
+        "words": 9529,
+        "contrast_frames": 4,
+        "claim_emphasis": 1,
+        "relation_markers": 84,
+      },
+      {
+        "paper": "rr-dataset-quality-acsac24",
+        "words": 8130,
+        "contrast_frames": 3,
+        "claim_emphasis": 0,
+        "relation_markers": 32,
+      },
+      {
+        "paper": "secure-acsac24",
+        "words": 7519,
+        "contrast_frames": 3,
+        "claim_emphasis": 0,
+        "relation_markers": 38,
+      },
+    ],
+  },
+
   "format": {
     // \textwidth and \textheight of [conference,compsoc], measured (ieee-conference.jsonc).
     "textblock_w_in": 7.0,
     "textblock_h_in": 9.02,
 
-    // 🔴 PAGE LIMITS ARE NOT GATED HERE, AND THAT IS A MEASUREMENT. banal splits pages into body
-    // and references by what is on them, and a page that holds ANY non-bibliography text counts as
-    // body. Measured 2026-09-27 on a [conference,compsoc] paper whose body ends at the foot of
-    // page 12:
+    // 🔴 THE PAGE LIMIT IS COUNTED BEFORE THE REFERENCES, NOT BY banal. banal splits pages into
+    // body and references by what is on them, and a page that holds ANY non-bibliography text
+    // counts as body. Measured 2026-09-27 on a [conference,compsoc] paper whose body ends at the
+    // foot of page 12:
     //   LLM Usage Statement before the references → body 13, references 0
     //   an appendix before the references         → body 13, references 2
     //   both placed after the references          → body right, the tail counted as references
     // AIDC counts neither in the limit («excluding references and appendices»; the ACSAC AI policy:
-    // the LLM Usage Statement «does not count towards the page limit»), and tex/required-section
-    // accepts the statement before or after the bibliography. A gate on banal's body count would
-    // therefore fail a correct 12-page paper — the realm.jsonc case again. The numbers below are
-    // data for humans, not an input to pdf/limits.
-    "page_limits_not_gated": {
-      "_": "Body pages, excluding references, appendices and the LLM Usage Statement. banal counts a page holding an appendix or the statement before the bibliography as body, so its body count is not this number.",
-      "regular": 12,
-      "short": 6,
-      "references_and_appendices": "unlimited",
-      "llm_usage_statement": "not counted (ACSAC AI policy)",
-    },
+    // the LLM Usage Statement «does not count towards the page limit»). So format/page-limit counts
+    // the body up to where the references start (`body_ends_at`): the page they start on counts
+    // when body text stands above the heading there, and the limit below is gated. An appendix
+    // ends the body too, wherever it sits, when hyperref anchors it (docs/rules/format/page-limit.md).
+    "body_ends_at": "references",
 
     // «Short position papers or work-in-progress (WIP) papers» share one limit, so one kind.
+    // References, appendices and the LLM Usage Statement: unlimited.
     "kinds": {
-      "regular": {},
-      "short": {},
+      "regular": { "body_pages_max": 12 },
+      "short": { "body_pages_max": 6 },
     },
   },
 }

package/presets/ieee-conference.jsonc

@@ -59,6 +59,14 @@
     "body_pt": 10,
     "body_pt_tol": 0.5,
 
+    // The class sets the bibliography in \footnotesize, 8 pt at 10pt and at 9pt (IEEEtran.cls,
+    // \thebibliography). Measured 2026-09-29 on the four built ACSAC papers in
+    // fixtures/accepted-papers (body-size.json): pdf.js reads the reference glyphs at 8.00 pt
+    // (7.97 without compsoc), banal gives 8.3 on all four. body_pt_tol widens it to 7.5-8.5, so a
+    // bibliography shrunk to \scriptsize (7.3) or grown to \small (9.3) is reported.
+    "ref_pt_min": 8,
+    "ref_pt_max": 8,
+
     // The Times clone. Headings and the title are set in the same family, so a `fonts_title` would
     // repeat this check and report a Computer Modern build twice.
     "fonts_text": "NimbusRomNo9L",

package/presets/venue-profile.schema.json

@@ -24,6 +24,10 @@
       "type": "string",
       "minLength": 1
     },
+    "blind": {
+      "description": "The venue reviews double-blind: its call says submissions must be anonymized. Read by `anonymity/identity`, which then requires the paper's `identity` and finds none of it in the PDF. Set only where the venue says so, with the quote. A child's value replaces its parent's.",
+      "type": "boolean"
+    },
     "aliases": {
       "description": "Other names the venue goes by in a paper's text — its acronym, its parent conference (`[\"AISec\", \"ACM CCS\"]`). Read by `tex/venue-leftover`: a paper that extends another venue and says one of these is flagged. Matched as whole words, case-sensitive. Collected along the `extends` chain, with `name`.",
       "type": "array",
@@ -105,6 +109,11 @@
           "type": "string",
           "minLength": 1
         },
+        "body_ends_at": {
+          "description": "How `format/page-limit` counts body pages. Absent: banal's body count (every page holding non-bibliography text). `references`: the pages up to where the body ends — at the references or the appendix, whichever comes first, found by hyperref's anchors and then the heading line — with that page counted when body text stands above the heading on it; for a venue whose limit excludes references and appendices (and what follows them, such as statements).",
+          "type": "string",
+          "enum": ["references"]
+        },
         "page_limits_not_gated": {
           "type": "object"
         },
@@ -128,8 +137,55 @@
       }
     },
     "rules": {
-      "description": "Rules the venue implies, rule id \u2192 ESLint entry (a severity, or [severity, options]). Applied to every paper that extends this preset, before the paper's own rules.",
+      "description": "Rules the venue implies, rule id → ESLint entry (a severity, or [severity, options]). Applied to every paper that extends this preset, before the paper's own rules.",
       "type": "object"
+    },
+    "register": {
+      "description": "The register rules' anchors (`tex/contrast-frames`, `tex/claim-emphasis`, `tex/relation-markers`): accepted papers of the venue, or of its parent conference, each with its body's word count and its count of each measure as paperlint's own counters give them. A measure's band is the anchors' range widened by two Poisson standard deviations at each end (docs/rules/tex/contrast-frames.md). A child preset's block replaces its parent's.",
+      "type": "object",
+      "additionalProperties": false,
+      "required": ["anchors"],
+      "properties": {
+        "anchors": {
+          "type": "array",
+          "minItems": 1,
+          "items": {
+            "type": "object",
+            "additionalProperties": false,
+            "required": [
+              "paper",
+              "words",
+              "contrast_frames",
+              "claim_emphasis",
+              "relation_markers"
+            ],
+            "properties": {
+              "paper": {
+                "description": "The anchor: a directory under fixtures/accepted-papers/ when the paper is there, or how to find it.",
+                "type": "string",
+                "minLength": 1
+              },
+              "words": {
+                "description": "Words of the body, as the register rules count them.",
+                "type": "integer",
+                "minimum": 1
+              },
+              "contrast_frames": {
+                "type": "integer",
+                "minimum": 0
+              },
+              "claim_emphasis": {
+                "type": "integer",
+                "minimum": 0
+              },
+              "relation_markers": {
+                "type": "integer",
+                "minimum": 0
+              }
+            }
+          }
+        }
+      }
     }
   },
   "allOf": [

package/skills/grade-paper-writing/SKILL.md

@@ -4,7 +4,7 @@ description: Use when asking "does this paper read well?" / "is the writing any
 allowed-tools: [Read, Write, Edit, Grep, Glob, Agent, Skill]
 ---
 
-<!-- vigiles:sha256:baf1f69df2502086 compiled from skills/grade-paper-writing/SKILL.md.spec.ts -->
+<!-- vigiles:sha256:03d0377baf71967f compiled from skills/grade-paper-writing/SKILL.md.spec.ts -->
 
 # grade-paper-writing — grade how the paper READS, then fix it sentence by sentence
 
@@ -172,16 +172,41 @@ is what happened.
 3. **For every dimension scoring ≤3, name the specific offending sentence or section** and write the fix —
    not "tighten the prose" but the rewritten sentence, or "move the point from ¶3 to ¶1", or "gloss
    `operation-normalized` in plain words at first use." Fixes must be applyable, not vibes.
-3a. **Grade #10, register distance, against the venue's own papers — not against the exemplars.**
-   Take 2–3 papers accepted at the target venue (or, for a first-edition workshop, at its parent
-   conference) in the same paper type: fetch them from the proceedings or the program page, or use the
-   ones the author supplies. Read the abstract, the introduction and the conclusion of each beside the
-   draft's, and record, for both sides: how each result is owned (*we measured* · a citation · a named
-   population), how often a point is made by contrast (*not X but Y*, *rather than*, *instead of* —
-   the measured difference between this corpus and accepted papers), sentence length and fragments,
-   and whether any sentence is an aphorism or a closing punchline. The
-   score is the distance: 5 = a reviewer could not tell the draft from the reference papers by register;
-   1 = it reads as a different genre. Quote the reference sentence beside each draft sentence you flag.
+3a. **Grade #10, register distance, against the venue's own papers, section by section — not against
+   the exemplars.** The rules count and this step judges; it starts from their numbers, not from a
+   count of its own.
+   - **Take the counts from the rules.** `npx paperlint lint <paper-dir>` reports
+     `tex/contrast-frames`, `tex/claim-emphasis` and `tex/relation-markers` when the venue preset
+     records anchors (AIDC does): each gives the body's rate per 10,000 words, the band, and the
+     anchors' range, and above the band it reports every frame and every claim in bold where it
+     stands. Those locations are the worklist. Do not recount them by grep — a narrower count is
+     gamed by rewording (*, not* → *rather than*), and the rules count every form.
+   - **Take the reference papers from the preset.** Its `register.anchors` names the accepted papers
+     the band comes from (AIDC: five ACSAC papers, all on arXiv under CC licences). Read two or three
+     of the same paper type as LaTeX source (`https://arxiv.org/e-print/<id>`) or as built PDFs. A
+     preset without anchors: fetch 2–3 accepted papers of the venue (or, for a first-edition workshop,
+     of its parent conference) from the proceedings or the program page, or use the ones the author
+     supplies.
+   - **Compare section by section, not the paper as a whole:** abstract, introduction, threat model or
+     background, method, results, limitations, conclusion. For each, put the draft's section beside the
+     same section of each reference paper and record, for both sides: how each result is owned (*we
+     measured* · a citation · a named population); how the relation between neighbouring sentences is
+     named (*because*, *therefore*, *however*, *for example*) or left to infer (a colon, a dash, a short
+     sentence); how often a point is made by what it is not; what is set in bold; and whether any
+     sentence is an aphorism, a fragment or a closing punchline. Quote the reference sentence beside
+     each draft sentence you flag.
+   - **Return concrete rewrites, and every one keeps every number and claim.** For each flagged sentence:
+     the draft sentence → the rewrite → the reference sentence it is modelled on → one line saying that
+     every number, citation and hedge is unchanged and the claim is neither softened nor strengthened.
+     A contrast frame is rewritten by saying what the thing is and naming the relation (*because*,
+     *as a result*), not by switching to another contrast form; a claim in bold loses the bold and
+     gains its owner (*we measured*, `\ref` to the table); a relation left to infer gets its word.
+     Where the frame IS the claim (the paper's thesis sets one mechanism against another), keep it
+     once, in the abstract and the introduction, and say so.
+   - **Score the distance:** 5 = a reviewer could not tell the draft from the reference papers by
+     register, section by section; 1 = it reads as a different genre. After the rewrites, run
+     `npx paperlint lint` again: every register rule inside its band is the floor for a 4, not the
+     score itself.
    If no accepted paper can be fetched or supplied, write `#10: not graded — no reference papers` and
    say which were tried; a register score from memory is the guess this dimension exists to replace.
 4. **Find the paper's three-move opportunities:** what held assumption could open it? what is its single
@@ -353,10 +378,13 @@ changes neither meaning nor register — and never rewrite a cited work's title
 **Signposting is the register, not a tic, and this skill never removes it on style grounds.** Frame
 markers and relation words — *we show*, *this section*, *first … finally*, *in contrast*, *therefore*,
 *however* — are how academic prose tells the reader what kind of claim comes next and how it relates to
-the last one. Their count does not separate a draft from accepted papers (see "Contrast framing"
-below); removing them does, in the wrong direction. Flag a connective only when it asserts a relation
-the two sentences do not have (a *therefore* that does not follow), or when it is empty filler (*it is
-worth noting that*). Never flag one for being academic.
+the last one. Relation words are a floor: `tex/relation-markers` reports a body below the band its
+venue's accepted papers set (AIDC: 25.4 per 10,000 words; the paper reviewers called a blog post stood
+at 19.1, and never wrote *however* once, where every accepted ACSAC paper writes it 6–11 times). Frame
+markers do not separate a draft from accepted papers in either direction (see "Contrast framing"
+below). Flag a connective only when it asserts a relation the two sentences do not have (a *therefore*
+that does not follow), or when it is empty filler (*it is worth noting that*). Never flag one for being
+academic.
 
 **Other markers worth a pass:** nominalisation density (*an honest accounting of what the construction
 costs* — abstract nouns doing the work verbs should do) and statistics chained through a paragraph
@@ -382,14 +410,15 @@ Two separable failures:
 
 ### 1. Contrast framing, and a frame that carries nothing
 
-**What separates this corpus from accepted papers is contrast, not signposting — measure that
-first.** Counted with the same counter on five accepted ACSAC 2024/25 ML-security papers against a
-paper of ours: *"not X but Y"* at 35.5 per 10,000 words in ours against at most 4.9 there (and five
-of their six hits were *"not only … but also"*); *"rather than / instead of"* at 19.9 against at most
-5.4. Every sentence that defines its point by what it is not is a sentence whose point the reader
-must reconstruct. **Rewrite heuristic:** state the thing; move the rejected alternative to its own
-sentence, or drop it if nobody proposed it. Check *instead of · rather than · not … but* first,
-against the same count on the reference papers of dimension #10.
+**What separates this corpus from accepted papers is contrast, not signposting — and a rule counts
+it.** `tex/contrast-frames` counts every form (*X, not Y* · *not X but Y* · *rather than* ·
+*instead of* · *as opposed to*) per 10,000 words against the band the venue preset's anchors set: the
+five accepted ACSAC papers measure 1.3–5.5, AIDC's band is 0–10.4, and the paper reviewers called a blog
+post measured 101 before its rewrite and 28 after it. Every sentence that defines its point by what
+it is not is a sentence whose point the reader must reconstruct. **Rewrite heuristic:** state the
+thing and name the relation; move the rejected alternative to its own sentence, or drop it if nobody
+proposed it. Swapping one form for another (*, not* → *rather than*) moves nothing — the rule counts
+both, and so does a reviewer.
 
 **Frame markers do not separate us from accepted papers, and are not a target in either direction.**
 Hyland's *frame markers* (*this paper · this section · we show · first … finally · in what follows*)
@@ -464,10 +493,13 @@ one did, and it was the worst section in the paper by measured density both time
 
 ### What is NOT mechanizable here
 
-Whether a frame marker carries its claim (*"we claim nothing there"* genuinely scopes one) and
-whether a hedge is honest or defensive. The mechanical leg that does exist: `tex/claim-provenance`
-reports every sentence of the body that holds a number and no owner. Count them, surface them, and leave the verdict to a reader.
-The counter's job is that nobody can say afterwards they did not know.
+Whether a frame marker carries its claim (*"we claim nothing there"* genuinely scopes one), whether a
+hedge is honest or defensive, whether a given contrast frame is the one a reader needs, and whether a
+rewrite kept every claim at its strength. The mechanical legs that do exist: `tex/claim-provenance`
+reports every sentence of the body that holds a number and no owner, and `tex/contrast-frames`,
+`tex/claim-emphasis` and `tex/relation-markers` report a body outside its venue's band. Count them,
+surface them, and leave the verdict to a reader. The counter's job is that nobody can say afterwards
+they did not know.
 
 ## Compose with
 - `harden-paper` — the multi-axis pre-submit gate calls this as its *writing* axis (the one a defect-hunt

package/skills/grade-paper-writing/SKILL.md.spec.ts

@@ -178,16 +178,41 @@ is what happened.
 3. **For every dimension scoring ≤3, name the specific offending sentence or section** and write the fix —
    not "tighten the prose" but the rewritten sentence, or "move the point from ¶3 to ¶1", or "gloss
    \`operation-normalized\` in plain words at first use." Fixes must be applyable, not vibes.
-3a. **Grade #10, register distance, against the venue's own papers — not against the exemplars.**
-   Take 2–3 papers accepted at the target venue (or, for a first-edition workshop, at its parent
-   conference) in the same paper type: fetch them from the proceedings or the program page, or use the
-   ones the author supplies. Read the abstract, the introduction and the conclusion of each beside the
-   draft's, and record, for both sides: how each result is owned (*we measured* · a citation · a named
-   population), how often a point is made by contrast (*not X but Y*, *rather than*, *instead of* —
-   the measured difference between this corpus and accepted papers), sentence length and fragments,
-   and whether any sentence is an aphorism or a closing punchline. The
-   score is the distance: 5 = a reviewer could not tell the draft from the reference papers by register;
-   1 = it reads as a different genre. Quote the reference sentence beside each draft sentence you flag.
+3a. **Grade #10, register distance, against the venue's own papers, section by section — not against
+   the exemplars.** The rules count and this step judges; it starts from their numbers, not from a
+   count of its own.
+   - **Take the counts from the rules.** \`npx paperlint lint <paper-dir>\` reports
+     \`tex/contrast-frames\`, \`tex/claim-emphasis\` and \`tex/relation-markers\` when the venue preset
+     records anchors (AIDC does): each gives the body's rate per 10,000 words, the band, and the
+     anchors' range, and above the band it reports every frame and every claim in bold where it
+     stands. Those locations are the worklist. Do not recount them by grep — a narrower count is
+     gamed by rewording (*, not* → *rather than*), and the rules count every form.
+   - **Take the reference papers from the preset.** Its \`register.anchors\` names the accepted papers
+     the band comes from (AIDC: five ACSAC papers, all on arXiv under CC licences). Read two or three
+     of the same paper type as LaTeX source (\`https://arxiv.org/e-print/<id>\`) or as built PDFs. A
+     preset without anchors: fetch 2–3 accepted papers of the venue (or, for a first-edition workshop,
+     of its parent conference) from the proceedings or the program page, or use the ones the author
+     supplies.
+   - **Compare section by section, not the paper as a whole:** abstract, introduction, threat model or
+     background, method, results, limitations, conclusion. For each, put the draft's section beside the
+     same section of each reference paper and record, for both sides: how each result is owned (*we
+     measured* · a citation · a named population); how the relation between neighbouring sentences is
+     named (*because*, *therefore*, *however*, *for example*) or left to infer (a colon, a dash, a short
+     sentence); how often a point is made by what it is not; what is set in bold; and whether any
+     sentence is an aphorism, a fragment or a closing punchline. Quote the reference sentence beside
+     each draft sentence you flag.
+   - **Return concrete rewrites, and every one keeps every number and claim.** For each flagged sentence:
+     the draft sentence → the rewrite → the reference sentence it is modelled on → one line saying that
+     every number, citation and hedge is unchanged and the claim is neither softened nor strengthened.
+     A contrast frame is rewritten by saying what the thing is and naming the relation (*because*,
+     *as a result*), not by switching to another contrast form; a claim in bold loses the bold and
+     gains its owner (*we measured*, \`\\ref\` to the table); a relation left to infer gets its word.
+     Where the frame IS the claim (the paper's thesis sets one mechanism against another), keep it
+     once, in the abstract and the introduction, and say so.
+   - **Score the distance:** 5 = a reviewer could not tell the draft from the reference papers by
+     register, section by section; 1 = it reads as a different genre. After the rewrites, run
+     \`npx paperlint lint\` again: every register rule inside its band is the floor for a 4, not the
+     score itself.
    If no accepted paper can be fetched or supplied, write \`#10: not graded — no reference papers\` and
    say which were tried; a register score from memory is the guess this dimension exists to replace.
 4. **Find the paper's three-move opportunities:** what held assumption could open it? what is its single
@@ -359,10 +384,13 @@ changes neither meaning nor register — and never rewrite a cited work's title
 **Signposting is the register, not a tic, and this skill never removes it on style grounds.** Frame
 markers and relation words — *we show*, *this section*, *first … finally*, *in contrast*, *therefore*,
 *however* — are how academic prose tells the reader what kind of claim comes next and how it relates to
-the last one. Their count does not separate a draft from accepted papers (see "Contrast framing"
-below); removing them does, in the wrong direction. Flag a connective only when it asserts a relation
-the two sentences do not have (a *therefore* that does not follow), or when it is empty filler (*it is
-worth noting that*). Never flag one for being academic.
+the last one. Relation words are a floor: \`tex/relation-markers\` reports a body below the band its
+venue's accepted papers set (AIDC: 25.4 per 10,000 words; the paper reviewers called a blog post stood
+at 19.1, and never wrote *however* once, where every accepted ACSAC paper writes it 6–11 times). Frame
+markers do not separate a draft from accepted papers in either direction (see "Contrast framing"
+below). Flag a connective only when it asserts a relation the two sentences do not have (a *therefore*
+that does not follow), or when it is empty filler (*it is worth noting that*). Never flag one for being
+academic.
 
 **Other markers worth a pass:** nominalisation density (*an honest accounting of what the construction
 costs* — abstract nouns doing the work verbs should do) and statistics chained through a paragraph
@@ -388,14 +416,15 @@ Two separable failures:
 
 ### 1. Contrast framing, and a frame that carries nothing
 
-**What separates this corpus from accepted papers is contrast, not signposting — measure that
-first.** Counted with the same counter on five accepted ACSAC 2024/25 ML-security papers against a
-paper of ours: *"not X but Y"* at 35.5 per 10,000 words in ours against at most 4.9 there (and five
-of their six hits were *"not only … but also"*); *"rather than / instead of"* at 19.9 against at most
-5.4. Every sentence that defines its point by what it is not is a sentence whose point the reader
-must reconstruct. **Rewrite heuristic:** state the thing; move the rejected alternative to its own
-sentence, or drop it if nobody proposed it. Check *instead of · rather than · not … but* first,
-against the same count on the reference papers of dimension #10.
+**What separates this corpus from accepted papers is contrast, not signposting — and a rule counts
+it.** \`tex/contrast-frames\` counts every form (*X, not Y* · *not X but Y* · *rather than* ·
+*instead of* · *as opposed to*) per 10,000 words against the band the venue preset's anchors set: the
+five accepted ACSAC papers measure 1.3–5.5, AIDC's band is 0–10.4, and the paper reviewers called a blog
+post measured 101 before its rewrite and 28 after it. Every sentence that defines its point by what
+it is not is a sentence whose point the reader must reconstruct. **Rewrite heuristic:** state the
+thing and name the relation; move the rejected alternative to its own sentence, or drop it if nobody
+proposed it. Swapping one form for another (*, not* → *rather than*) moves nothing — the rule counts
+both, and so does a reviewer.
 
 **Frame markers do not separate us from accepted papers, and are not a target in either direction.**
 Hyland's *frame markers* (*this paper · this section · we show · first … finally · in what follows*)
@@ -470,10 +499,13 @@ one did, and it was the worst section in the paper by measured density both time
 
 ### What is NOT mechanizable here
 
-Whether a frame marker carries its claim (*"we claim nothing there"* genuinely scopes one) and
-whether a hedge is honest or defensive. The mechanical leg that does exist: \`tex/claim-provenance\`
-reports every sentence of the body that holds a number and no owner. Count them, surface them, and leave the verdict to a reader.
-The counter's job is that nobody can say afterwards they did not know.
+Whether a frame marker carries its claim (*"we claim nothing there"* genuinely scopes one), whether a
+hedge is honest or defensive, whether a given contrast frame is the one a reader needs, and whether a
+rewrite kept every claim at its strength. The mechanical legs that do exist: \`tex/claim-provenance\`
+reports every sentence of the body that holds a number and no owner, and \`tex/contrast-frames\`,
+\`tex/claim-emphasis\` and \`tex/relation-markers\` report a body outside its venue's band. Count them,
+surface them, and leave the verdict to a reader. The counter's job is that nobody can say afterwards
+they did not know.
 
 ## Compose with
 - \`harden-paper\` — the multi-axis pre-submit gate calls this as its *writing* axis (the one a defect-hunt

package/skills/render-paper/extract-pdf-facts.harness.ts

@@ -119,8 +119,8 @@ try {
   const fonts = Fonts.safeParse(facts.fonts);
   const fontList = fonts.success ? fonts.data : [];
   check(
-    "the facts are schema 2, about paper.pdf, for the venue paperlint.json declares",
-    facts.schema === 2 &&
+    "the facts are schema 3, about paper.pdf, for the venue paperlint.json declares",
+    facts.schema === 3 &&
       facts.pdf === "paper.pdf" &&
       facts.venue === "agenticdev" &&
       facts.kind === "short",
@@ -236,5 +236,5 @@ try {
 }
 
 console.log(
-  `✓ ${String(check.count)} assertions passed — extract-pdf-facts: the exit-code contract, schema 2, and the no-no-yes font`,
+  `✓ ${String(check.count)} assertions passed — extract-pdf-facts: the exit-code contract, schema 3, and the no-no-yes font`,
 );