references/forbidden-patterns.md
# Forbidden patterns
## Table of contents
1. Em-dash misuse
2. Banned AI-tone vocabulary
3. Chartjunk patterns
4. Overclaim patterns
5. Plagiarism red lines
6. Detection heuristics
## 1. Em-dash misuse
Em-dashes used as sentence connectors are banned in this
project's writing conventions. Two common misuse patterns:
### Misuse A: connecting two independent clauses
- Bad: "... closes the research loop, from literature analysis to
deployment."
- Good (using comma plus subordinate clause): "... closes the
research loop, covering literature analysis, hypothesis
generation, simulation, validation, and deployment."
### Misuse B: inserting a parenthetical
- Bad: "This approach, though simple, is highly effective."
- Good (using a subordinate clause): "Although simple, this
approach is highly effective."
Acceptable em-dash uses are rare in academic writing. When in
doubt, do not use them. Commas, colons, or periods handle nearly
every case.
Detection: grep for the Unicode em-dash character. Every
occurrence in the body is a MAJOR finding unless explicitly in a
code sample or preserved quotation.
## 2. Banned AI-tone vocabulary
The following vocabulary signals AI-authored or AI-assisted prose
to experienced reviewers and should be avoided. Seeing three or
more uses in a paper is a MAJOR finding.
### Innovation hyperbole
innovative, pioneering, revolutionary paradigm, transformative
framework
### Performance hyperbole
superior, surpass, excel, remarkable, unprecedented, achieves
SOTA, breakthrough performance
### Contribution-summary markers
general-purpose, is capable of
### Logical connective markers
notably, yet, yielding, at its essence
### Overused verbs
encompass, differentiate, reveal, underscore, surpass, exhibit
superior capability, exceed, pave the way for, highlight the
potential of
### Overused phrases
profound challenges (this collocation does not exist in standard
writing), stems from
### Overused adjectives and verbs
rigid, impede
Replace with neutral technical verbs from the section-guides
reference: propose, introduce, design, present, show, demonstrate,
report, observe.
## 3. Chartjunk patterns
Chart elements that add visual noise without information:
- 3D bar charts, 3D pie charts.
- Drop shadows on bars or points.
- Gradient fills except where a gradient encodes a continuous
variable.
- Heavy grid lines.
- Cross-hatching for decoration.
- Chart titles duplicating the caption.
- Legend keys with more than six entries (try grouping instead).
Each chartjunk instance is a MINOR or MAJOR finding depending on
severity.
## 4. Overclaim patterns
Overclaims common in draft papers:
- "Our method is state-of-the-art" without qualification. State
state-of-the-art on which benchmark, under which conditions.
- "Comprehensive experiments" as a contribution. Experiments are
expected; what was specifically shown?
- "Extensive analysis". Name what the analysis found, not that
analysis exists.
- "We solve the problem of X". Solving is a strong claim; usually
the paper improves on X rather than solves.
- "We are the first to". Verify carefully; if the claim is wrong,
reviewers will find and quote it.
Each overclaim is a MAJOR finding. Reviewers keep a running
grudge against papers that overclaim in the Introduction; the
cost compounds over the review cycle.
## 5. Plagiarism red lines
Never copy sentences from other papers, including the author's
own prior papers without explicit reuse acknowledgement. This
applies to:
- Related Work summaries of prior methods.
- Introduction references to prior work.
- Method descriptions that resemble baseline papers' own
descriptions.
Plagiarism detection is a CRITICAL finding that blocks submission.
When the paper's Related Work or Introduction paraphrases prior
work too closely, rewrite in the author's own words.
## 6. Detection heuristics
For the banned-vocabulary scan:
- Case-insensitive substring search for each banned term.
- Count occurrences per term.
- Three or more occurrences of any single banned term is MAJOR.
- Single occurrence is MINOR.
For em-dash scan:
- Search for the Unicode em-dash character.
- Every occurrence is a finding.
- Severity is MAJOR if in body; MINOR if in preserved quotations.
For chartjunk scan:
- Cannot be done automatically; the skill relies on the user's
figure description or on an uploaded figure.
For overclaim scan:
- Look for unconditional superlatives: "is state-of-the-art", "is
superior", "beats all".
- Flag each occurrence with a request to qualify or substantiate.
For plagiarism:
- Random sentence-level search of the Related Work and
Introduction against public texts is not feasible inside the
skill. Recommend an external plagiarism checker for final
submission.
references/grammar-rules.md
# Grammar rules for non-native English authors
## Table of contents
1. Article usage
2. Subject-verb agreement
3. Tense consistency
4. Sentence complexity
5. Which versus that
6. Passive voice
7. Chinglish patterns
8. Punctuation in quotations
## 1. Article usage
Rule G1: singular countable nouns require an article (a, an, the)
unless they are generic plurals or proper nouns.
| Bad | Good |
|---|---|
| We propose novel method. | We propose a novel method. |
| in descending order | in a descending order |
| Our model supports the case changed. | Our model supports the case that changed. |
Quick test: if the noun is countable and singular, and the
sentence is about a specific instance, the article is required.
For generic plurals, articles are omitted. For uncountable nouns
(information, evidence), articles are usually omitted.
## 2. Subject-verb agreement
Rule G2: third-person singular subjects take third-person
singular verbs.
- it predicts, it proposes, it improves.
- we propose, we show, we introduce.
- they demonstrate, they report.
The ICML or VLDB paper (singular) uses the third-person singular
verb form:
- The method predicts accurately.
- Not: The method predict accurately.
## 3. Tense consistency
Rule G3: different sections have different default tenses.
- Abstract: present tense for the paper, past tense for
experiments.
- Introduction: present for the paper's claims, past for related
work references.
- Methodology: present tense (the method does X).
- Experiments: past tense (we ran the experiment, we observed).
- Related Work: past tense for summarising prior work's
contributions.
- Conclusion: present perfect tense (we have shown, we have
proposed).
Mixing tenses within a single sentence or paragraph is a common
reviewer complaint.
## 4. Sentence complexity
Rule G4: one sentence, one main idea. Two verbs in one sentence
require a connector. Avoid sentences with two independent clauses
joined only by a comma (comma splice).
| Bad | Good |
|---|---|
| We propose a method has high efficiency. | We propose a method that has high efficiency. |
| We propose a method has high efficiency. | We propose a high-efficiency method. |
| We propose a method, it has high efficiency. | We propose a method. It has high efficiency. |
| We propose a method, it has high efficiency. | We propose a method with high efficiency. |
Long sentences should be split at natural boundaries. Use
"Specifically," or "In particular," to start the continuation
sentence rather than stuffing everything into one complex
sentence.
## 5. Which versus that
Rule G5: use "that" for restrictive clauses (essential to the
meaning); use "which" for non-restrictive clauses (parenthetical
and set off by commas).
- Restrictive: "the method that achieves state of the art" (out
of a set, the one that achieves).
- Non-restrictive: "the method, which achieves state of the art,"
(an aside).
Non-native authors commonly use "which" in both cases. Correct
to "that" when the clause is restrictive.
## 6. Passive voice
Rule G6: prefer active voice unless the subject is genuinely not
important. "The method was evaluated" is acceptable when the
evaluator's identity does not matter; "We evaluated the method"
is stronger when it does.
Over-use of passive voice makes prose feel evasive. Audit
paragraphs for strings of passive constructions and replace the
most important sentences with active voice.
## 7. Chinglish patterns
Common Chinglish patterns to flag:
- Direct translation of Chinese idioms that do not carry in
English.
- Overuse of "very" (赶很多中文用户会用 "very"; often unnecessary).
- Use of the same word several times in short succession where
a synonym would be clearer.
- Over-hedged statements ("it may be the case that perhaps the
method possibly") when a direct statement is clearer.
- Translating a Chinese draft with a machine translator. Never do
this; the output accumulates un-idiomatic phrasing that
reviewers notice immediately. Write in English from the start.
## 8. Punctuation in quotations
Rule G8: in an academic writing context, use logical punctuation
(punctuation inside quotation marks only if part of the quoted
material). LLM-generated text often defaults to American typography
(punctuation always inside); correct this for academic English.
| American (avoid) | Logical (use) |
|---|---|
| approach is called "data-centric AI," which emphasises | approach is called "data-centric AI", which emphasises |
| outputs what it calls "semantic summaries." | outputs what it calls "semantic summaries". |
Also: use backticks and single or double quotation marks per
LaTeX convention: `` `` `` for open and `'' '` for close, not the
straight typewriter character.
references/latex-rules.md
# LaTeX format rules
## Table of contents
1. File organisation
2. Macro definitions
3. Citations
4. Labels and references
5. Figures and tables
6. Equations
7. Quotation marks and dashes
8. Revision conventions
## 1. File organisation
A multi-section paper should split content into files under a
`sections/` directory, with `main.tex` using `\input{sections/...}`.
Typical structure:
```
sections/
figs/
exps/
algs/
main.tex
commands.tex
bibfile.bib
```
A monolithic `main.tex` with everything inline is acceptable for
very short papers but harder to version-control across multiple
authors.
## 2. Macro definitions
Define macros for recurring strings: system names, method names,
commonly misspelled words, or frequently used highlighting.
Example in `commands.tex`:
```latex
\newcommand{\sys}{\texttt{Alpha-SQL}\xspace}
\newcommand{\hi}[1]{\vspace{.25em}\noindent \textbf{#1}}
\newcommand{\lgl}[1]{\textcolor{blue}{LGL: #1}}
\newcommand{\revision}[1]{\textcolor{blue}{#1}}
```
Renaming becomes a one-line change rather than a find-and-replace
across dozens of sections. When a system name changes mid-
revision, the macro is the only edit required.
## 3. Citations
Rule L1: always use the non-breaking tilde between a word and its
citation, to prevent awkward line breaks.
| Bad | Good |
|---|---|
| `ResNet\cite{X}` | `ResNet~\cite{X}` |
| `ResNet \cite{X}` | `ResNet~\cite{X}` |
Multiple citations go inside one command:
```latex
Artificial Intelligence~\cite{xxxx, yyyy, zzzz}
```
Rule L2: follow the venue's citation style. ACL uses natbib with
`\citep` for parenthetical and `\citet` for textual. Data
management venues (SIGMOD, VLDB) use IEEE or ACM numeric styles.
Convert before submission.
Rule L3: bib entries must include conference name, year, pages,
authors, title. Pull from DBLP for correctness.
## 4. Labels and references
Rule L4: labels use underscores and a prefix for readability.
- `\label{fig:system_overview}` (good).
- `\label{sec:intro}` (good).
- `\label{system overview}` (bad; contains a space).
- `\label{system-overview}` (bad; hyphens break some references).
Rule L5: references also use the non-breaking tilde.
```latex
as shown in Figure~\ref{fig:system_overview}
discussed in Section~\ref{sec:intro}
```
## 5. Figures and tables
Rule L6: every figure and table must have a caption. Captions
describe the finding, not just the setup, especially for
experimental figures.
Rule L7: default placement is top of page (`[t!]`). If the figure
does not fit at the top, `[b!]` (bottom) is acceptable. Inline
`[h]` placement rarely works; let LaTeX decide.
Rule L8: figures use vector formats (PDF, EPS, SVG). Never insert
PNG or JPG screenshots in the final paper; pixelation appears on
zoom.
Rule L9: always reference the figure or table in prose before it
appears. "Figure 2 shows..." is the canonical pattern.
## 6. Equations
Rule L10: every numbered equation should be referenced in the
prose at least once. If an equation is not referenced, either
reference it or remove the numbering with `\begin{equation*}`.
Rule L11: equation numbering should be contiguous within a
section. Use `\label` on each numbered equation to allow
rearranging without renumbering manually.
## 7. Quotation marks and dashes
Rule L12: use the LaTeX convention for quotation marks.
- Double quotes: use `` `` `` for open, `` '' `` for close.
- Single quotes: use `` ` `` for open, `` ' `` for close.
- Never use the straight ASCII typewriter `"` character.
Rule L13: dashes have three types in LaTeX; use each correctly.
- Hyphen `-`: compound adjectives (high-efficiency, zero-shot).
- En-dash `--`: number ranges (pages 10--15, 2024--2026).
- Em-dash `---`: semantic break. Project rule: do not use em-
dashes in the body. Use commas, colons, or periods instead.
## 8. Revision conventions
For revision submissions, preserve the review history in
dedicated files.
```
responses/
meta.tex
r1.tex
r2.tex
r3.tex
```
Mark revised text with `\revision{...}` (defined as a blue color
macro). Mark revision pointers in the margins with `\marginpar`:
```latex
\marginpar[]{\revision{R1.W1}}{\revision{text}}
\marginpar[\revision{R3.W2}]{}{\revision{text}}
```
Each `r<n>.tex` file opens with a summary table of the reviewer's
comments and the sections that respond to them. Make the summary
table the first content in the response file so reviewers can see
at a glance what was done.
references/logic-and-structure.md
# Logic and structure rules
## Table of contents
1. Logic First rule
2. Self-contained rule
3. Leading Text rule
4. Running Example rule
5. Paragraph discipline
6. Section-level consistency
7. Failure patterns
## 1. Logic First rule
Every sentence has a purpose. Every two consecutive sentences have
a connection. Every claim carries evidence. Every critique cites
a ground.
When reviewing a paragraph, ask for each sentence: what is this
sentence trying to do? If the answer is vague, the sentence is a
candidate for cutting or rewriting. When reviewing two consecutive
sentences, ask: how does sentence 2 advance the point made by
sentence 1? If the link is not apparent, a transitional phrase or
a restructure is needed.
Logic First is the highest-priority writing rule. A paper can
recover from many writing weaknesses if its logic is airtight;
strong prose cannot compensate for broken logic.
## 2. Self-contained rule
The paper should be readable without referencing external material
for basic understanding. A reader without access to prior work
should be able to follow the paper's contribution; external
citations are for context and credit, not for load-bearing
content.
Self-containment checks:
- Every abbreviation defined on first use.
- Every symbol defined on first use.
- Every cited method explained in one sentence before use.
- Figure captions readable without the body text.
## 3. Leading Text rule
Every framework section, every subsection, every paragraph should
begin with a leading sentence that says what the reader is about
to read. The leading text is the reader's navigation anchor.
Leading text does three things:
- States the goal of the upcoming content.
- Previews the structure (if the section or paragraph has
internal structure).
- Connects to what came before.
A paper without leading texts reads as a succession of claims
without context. Reviewers sense this as disorganisation.
## 4. Running Example rule
The running example established in Introduction Paragraph 1 should
reappear:
- In the Methodology as a walkthrough of how the method handles
it.
- In the Experiments as a case study or example in an analysis
figure.
- In the Conclusion optionally, as a closing reference.
A running example that appears only in the Introduction signals
that the author did not structure the narrative around a single
concrete anchor; reviewers struggle to follow the paper's story.
## 5. Paragraph discipline
Each paragraph:
- Opens with a topic sentence that states the paragraph's point.
- Contains three to eight sentences; significantly more usually
indicates the paragraph is trying to cover two topics.
- Closes with a sentence that either concludes the point or
bridges to the next paragraph.
- Contains only one topic; do not mix unrelated points in one
paragraph.
Common paragraph failures:
- No topic sentence; the point only emerges in the middle.
- Topic sentence buried; the reader must infer the point.
- Too long; the paragraph covers two topics that should be split.
- Too short; a two-sentence paragraph rarely earns its own line
break, fold into a neighbour.
## 6. Section-level consistency
Beyond the paragraph level, sections are expected to follow the
framework laid out in the Introduction:
- Methodology sections mention the module names introduced in
the Introduction.
- Experiments evaluate the claims made in the Contributions.
- Related Work is consistent with the Limitations cited in the
Introduction.
Inconsistency between sections is among the most common review
findings. Check that every module name in the Introduction appears
verbatim as a subsection title in the Methodology. Check that
every contribution cited in the Introduction has a corresponding
experiment that validates it.
## 7. Failure patterns
| Pattern | Detection | Severity |
|---|---|---|
| Broken flowchart (Introduction) | One of the six paragraphs missing | CRITICAL |
| Orphan contribution | Contribution has no section | CRITICAL |
| Running example abandoned after Introduction | Search the Methodology and Experiments for the example; not found | MAJOR |
| Leading text missing | A section or subsection opens with a dense paragraph | MAJOR |
| Topic sentence absent | Random paragraph sampled; the point is not in the first sentence | MAJOR |
| Module name mismatch | Methodology subsection titles differ from Introduction module names | MAJOR |
| Related Work contradicts Limitations | Methods criticised in Introduction appear without critique in Related Work | MAJOR |
| Paragraph over 10 lines | Length check | MINOR |
| Paragraph under 2 sentences | Length check | MINOR |
references/section-guides.md
# Per-section writing guides
## Table of contents
1. Abstract
2. Introduction
3. Problem Formulation
4. Framework or Methodology
5. Experiments
6. Related Work
7. Conclusion
## 1. Abstract
Five-sentence formula.
| Sentence | Content |
|---|---|
| 1 | What problem does the paper solve? Context plus significance |
| 2 | Why is it worth solving? Nobody has done it, or existing methods fail |
| 3 | What are the challenges? Must be concrete; challenges of the problem, not of the method |
| 4 | How is the problem solved? Method name and the key components, matched to the challenges |
| 5 | How well does the method perform? Real datasets, concrete numbers, code link if available |
Common failures:
- Too abstract on what and why.
- Missing challenges sentence.
- Results sentence without specific numbers.
- Hype words (revolutionary, groundbreaking) in the first
sentence; these read as defensive.
## 2. Introduction
Six-paragraph formula matching the Introduction Flowchart.
| Paragraph | Content |
|---|---|
| 1 | The problem; a concrete motivating example |
| 2 | Existing approaches and their limitations; highly condensed, at most three limitations |
| 3 | The problem's essence and hard constraints; the paper's goal or key idea |
| 4 | Key challenges; the challenges of the problem, not the method |
| 5 | Solution overview; one-to-one mapping with challenges |
| 6 | Summary of contributions, three or four, each mapped to a section |
Key constraints:
- Attracts the reviewer. Mostly decides acceptance.
- Highlights novelty and technical depth.
- States the difference from existing work explicitly.
- Writes the paper's own ideas; other work belongs in Related
Work.
- Terms are precise; the reader should grasp the paper's core
quickly.
## 3. Problem Formulation
Three steps:
1. Describe the problem in prose so the reader understands the
scope.
2. Present the formal definition with notation and equations.
3. Use a concrete example to anchor understanding.
Key constraint: a paper should solve one problem, not several.
A strong running example is worth more than additional formal
machinery.
## 4. Framework or Methodology
Organising principle: overall-to-detail. The section starts with
an architecture and then drills into components; each component's
subsection repeats the pattern.
| Part | Content |
|---|---|
| Architecture overview | Overall picture; after reading this part, the reader understands the technique at a high level |
| Workflow walkthrough | Traces inputs through the system end-to-end |
| Simple components | Described in-line in the Framework section |
| Complex components | Covered at high level here; full detail deferred to a dedicated subsection or section |
Must-have:
- Novelty and contribution explicitly stated.
- Overall architecture figure.
- If a component has no novelty, minimise coverage; long
descriptions of known techniques add noise.
- Later subsections order by novelty: strongest first.
- Self-contained; well-illustrated; leading text per subsection.
Avoid:
- Explaining why alternative methods were rejected in detail;
that belongs in Related Work or an ablation.
- Prior-work notation fragments mixed with the paper's notation.
- Symbol drift across subsections; define once and use
consistently.
## 5. Experiments
Tense: past.
Setup requirements:
- Experimental goals stated first.
- At least three datasets.
- Evaluation metrics named and justified.
- Baselines selected carefully; obvious baselines must not be
omitted. If the paper cites work proving A beats B and C, only A
is required, with explanation.
Overall comparison:
- End-to-end comparison against all baselines first.
- State the metric before analysing results.
- State the gain, explain the reason, extract findings.
Ablation study:
- Evaluate the contribution of each proposed component.
- For three components A, B, C: report All, -A, -B, -C.
- Vary key hyperparameters.
Must-have:
- Real datasets and persuasive baselines.
- Conclusions stated explicitly. What did the experiments
demonstrate? Why?
## 6. Related Work
Position:
- Section 2 if the paper depends on prior frameworks, models, or
methods for understanding.
- Penultimate section if the paper is self-contained. The
penultimate position is the preferred default for strong papers.
Do:
- Summarise in the author's own words. No copying of sentences.
- Fair crediting; do not over-criticise.
- Focus on high-level summary, not minor details.
- Self-contained; explain how the paper differs from related
work.
- Group by method category, then state differences. Criticise
closest-related work directly; less-related work gets less
critique and more acknowledgement.
- Citations use conference abbreviations (SIGMOD, VLDB, ICDE,
TKDE). Pull bibtex from DBLP.
Do not:
- Copy sentences from related work, ever. Plagiarism is a red
line.
## 7. Conclusion
Tense: present perfect (we have shown, we have proposed).
Purpose: summarise contributions and significance. Do not duplicate
the Introduction.
- Introduction addresses readers who have not yet read the paper.
- Conclusion addresses readers who have finished the paper; they
know the technical details.
Close with implications or future directions in one to two
sentences. Avoid vague claims about "impact".
SKILL.md
---
name: pre-submission-reviewer
description: >-
Runs a pre-submission review of a technical paper across five
dimensions: macro logic, writing details, English grammar, LaTeX
formatting, and figure quality. Uses a reviewer-style severity
taxonomy (CRITICAL / MAJOR / MINOR) and flags banned AI-tone
vocabulary and em-dash misuse. Use when the user asks to 'review
this paper', 'audit before submission', 'check the draft', 'find
issues', 'proofread', or within one week of a submission
deadline.
license: CC-BY-4.0
---
# Pre-Submission Reviewer
## Overview
Three to five days before a submission deadline is the window where
a careful external review pays off most. This skill takes a full
paper or key sections and produces a structured review across five
dimensions, each with severity-tagged findings and concrete
rewrite suggestions. It enforces the mechanical rules from the
writing-checklist section (no em-dashes, no banned AI-tone
vocabulary, leading text per paragraph, topic-sentence discipline,
citation-format uniformity) and surfaces the patterns that
non-native English-speaking authors most commonly violate
(articles, subject-verb agreement, tense consistency, which versus
that, Chinglish phrasing).
The output is not a rewrite. It is a prioritised list of findings
with severity tags; the author decides which to fix. CRITICAL items
should block submission until addressed.
## When to use this skill
- Three to five days before a submission deadline.
- The user asks to 'review this paper', 'audit before submission',
'check the draft', 'find issues', 'proofread'.
- After a camera-ready revision, before sending the final version.
- After any major rewrite (rebuttal responses, Section 3 overhaul).
- When the user suspects AI-tone contamination in a section.
## When NOT to use this skill
- The paper is still being structured. Use `tech-paper-template`,
`intro-drafter`, or `benchmark-paper-template` (separate plugin) first.
- The user wants structural advice rather than review. Use the
drafting skills instead.
## Core procedure
### Step 0: paradigm and venue fit
Before the dimensional review, settle two things.
**Paradigm**, judged by research method, because what counts as a
severe problem differs by paradigm:
- **STEM / technical** (CS, engineering, materials, chemistry): weight
the Introduction chain, contribution-to-section mapping, baseline
coverage, ablation-based attribution, figure quality. Benchmark
papers additionally get coverage, reproducibility, and
contamination checks; chemistry and materials get characterization
completeness, purity, controls, replicates.
- **Humanities** (literature, history, philosophy, discourse
analysis): weight whether the thesis is explicit and defensible,
core concepts pinned down and stable across sections, the
literature genuinely engaged, and the material (texts, archives,
cases) able to carry the claims. Do **not** demand baselines or
ablations here; check instead whether sub-arguments build on each
other rather than sit in parallel, and whether conclusions outrun
the material.
- **Empirical social science**: weight operationalized research
questions, sampling and data-source justification, statistics
matched to data types, conclusions bounded by the sample. For
theory papers, "experiments" reads as "proofs": check the proofs,
and never call the evidence thin merely because there is no results
table.
- **Finance / economics**: weight identification credibility,
endogeneity handling, robustness checks (their absence is a
first-round flag), and economic versus merely statistical
significance.
- **Law**: weight the accuracy of statutes, case numbers, and
holdings (an invented or wrong citation is rejection-level), the
clarity of the interpretive approach, and whether comparative
arguments state their scope.
When the paradigm is unclear, ask the author before reviewing with
the wrong ruler.
**Venue fit**: if the stated target venue's scope visibly mismatches
the paper's topic or contribution type, that is a real rejection
risk, not a taste note. Flag it as a finding and suggest two or three
better-fitting venues.
### Step 1: Dimension 1 Macro logic review
See: references/logic-and-structure.md for the Logic First rule,
Self-contained rule, Leading Text rule, and Running Example rule.
Check:
- Introduction flowchart is intact (Background, Limitations, Goal
or Key Idea, Challenges, Methodology, Contributions).
- Contributions map one-to-one with methodology modules and with
section numbers.
- Experiments validate the paper's main claims, not tangential
ones.
- Related Work covers the necessary prior art.
- Running example is consistent across Introduction, Methodology,
Experiments.
- The headline result's attribution is isolated: an ablation
separates the core mechanism from peripheral factors (a routing
step, post-processing, a stronger base model, favorable samples).
No such ablation: flag "attribution unverified" as MAJOR.
- Claims match their evidence: "solves" is stronger than most papers
earn (usually "improves"); "state-of-the-art" needs the benchmark
and conditions; "we are the first" gets checked or flagged.
Every break in the chain is CRITICAL.
**Retrieval-grounded checks** (when the environment has a
literature-search capability: a scholarly tool, web search over
scholarly indexes, or shell access to public APIs):
1. **Novelty verification**: extract two or three keyword groups from
the paper's core method and problem setting, retrieve, and identify
the three to five closest published works. If Related Work already
covers them, the paper's positioning stands; a highly relevant
uncovered work is a MAJOR finding ("missed X, Author et al.,
Year"). Compare on difference axes; a similar title alone proves
nothing.
2. **Citation completeness**: retrieve the field's recent
representative works and its canonical ones, and compare against
the reference list; a missing canonical baseline, founding paper,
or recent survey is MAJOR.
Retrieval results support metadata-level judgments only; never quote
numbers or method details from search snippets. Without any retrieval
capability, skip these two checks and say so in the summary.
### Step 2: Dimension 2 Writing details review
See: references/logic-and-structure.md for paragraph-level rules.
Check:
- Every paragraph has a topic sentence.
- Paragraphs transition smoothly; no orphan paragraphs.
- Paragraphs are not over 10 lines; split if so.
- No repeated or redundant passages.
- Abstract covers problem, method, result.
### Step 3: Dimension 3 English grammar review
See: references/grammar-rules.md for the canonical list of errors
common to non-native English authors, with corrections and
examples.
Check the usual suspects:
- Article use (a, an, the).
- Subject-verb agreement (third-person singular).
- Tense consistency (Related Work past, method present).
- Passive-voice overuse.
- Which versus that.
- Sentence length; split long sentences at "Specifically,".
- Chinglish patterns.
### Step 4: Dimension 4 LaTeX format review
See: references/latex-rules.md for the canonical list of LaTeX-
specific issues.
Check:
- Equation numbering contiguous; every numbered equation
referenced.
- Figures and tables have captions; captions are detailed.
- Citations use the correct command and the non-breaking tilde
(for example, `ResNet~\cite{X}`, never `ResNet\cite{X}`).
- Labels use underscores, not spaces or hyphens.
- Vector figure format; no raster.
- Page-limit compliance.
### Step 5: Dimension 5 Figure quality review
See: references/forbidden-patterns.md for chartjunk patterns and
the full figure-quality checklist.
For each figure:
- Vector format.
- Font size large enough post-scaling.
- Colour-blind-safe palette; dual encoding.
- Self-contained caption with a finding in the first sentence.
- No chartjunk.
- Motivated example is concrete and failure-revealing.
- Solution overview has labels matching section titles.
### Step 6: Banned-vocabulary and em-dash scan
See: references/forbidden-patterns.md for the banned-word list.
Scan the full paper for:
- Em-dashes used as sentence connectors (banned; project rule).
- AI-tone words: innovative, pioneering, revolutionary paradigm,
transformative framework, superior, surpass, excel, remarkable,
unprecedented, breakthrough performance, general-purpose, is
capable of, notably, yet, yielding, at its essence, encompass,
differentiate, reveal, underscore, pave the way for, highlight
the potential of, profound challenges, stems from, rigid,
impede.
Flag each occurrence with a severity tag. Em-dashes are MAJOR by
default; banned AI-tone words are MAJOR if they appear three or
more times.
### Step 7: Section-by-section review
See: references/section-guides.md for the per-section writing
guides for Abstract, Introduction, Problem Formulation, Framework
or Method, Experiments, Related Work, and Conclusion.
For each section, check that the section's content matches the
guide's canonical structure (for example, Abstract's five-sentence
formula: what, why, challenges, how, results).
### Step 8: Integrity gate
Run the checks in the Integrity gate section below.
### Step 9: Output
Emit the review in the Output format below.
## Severity taxonomy
- **CRITICAL**: blocks submission. Example: contributions do not
map to sections; introduction flowchart broken; no real-world
running example; raster figure in final draft; missing key
baseline; page-limit violation.
- **MAJOR**: reviewers will flag in first round. Example:
topic-sentence absent from 3+ paragraphs; em-dash in 5+ places;
banned AI-tone word in 3+ places; Table 1 comparison missing;
chart type mismatched with data.
- **MINOR**: polish. Example: two long sentences that could be
split; default Matplotlib styling; single article error.
Severity honesty cuts both ways. A review that lists a dozen MINOR
items while missing the one rejection-level flaw sends the author to
submission with false confidence; a review that inflates taste issues
into CRITICAL destroys trust. The overall recommendation must match
the findings: any unresolved CRITICAL forbids "ready to submit", and
a near-ready verdict requires zero CRITICAL and at most two MAJOR.
## Integrity gate
Each bullet is tagged [inspection] (LLM verifies from the paper
text) or [attestation] (LLM runs the procedure and states it has
done so; user remains responsible for confirming completeness).
Before emitting the review:
1. **[inspection]** Every finding quotes specific text (sentence,
phrase, figure name); no "the Introduction is unclear" without
a quoted line.
2. **[inspection]** Every CRITICAL finding has a concrete fix
suggestion, not "rewrite entirely".
3. **[inspection]** No fabricated quotes: only text actually
present in the submitted material.
4. **[inspection]** Severity assignments follow the taxonomy;
nothing is marked CRITICAL for taste reasons.
5. **[inspection]** Dimension 3 (grammar) findings cite the
specific grammar rule from `references/grammar-rules.md`.
6. **[attestation]** Dimension 6 banned-vocabulary scan is run in
full on the entire paper, not sampled. The skill attests the
full scan; if the paper is extremely long, the skill states it
chunked the input and describes the chunking strategy.
7. **[inspection]** Final score matches the CRITICAL + MAJOR
count; a score of 9 or 10 requires zero CRITICAL and at most
two MAJOR items.
If any [inspection] check fails, mark the output as "needs user
attention". For [attestation] bullets, the skill states the scope
of its scan and the user confirms completeness.
Run the gate silently. Do not print a per-gate pass or fail report;
a failure surfaces as a concrete finding in the affected dimension,
and the delivered review stays free of internal checking rituals.
## Output format
### Summary
- CRITICAL: <n>
- MAJOR: <m>
- MINOR: <k>
- Top three fixes first: ...
### Dimension 1: Macro logic
| # | Finding | Severity | Suggested fix |
|---|---|---|---|
| 1 | <quoted text> | CRITICAL or MAJOR or MINOR | <fix> |
### Dimension 2: Writing details
<same table shape>
### Dimension 3: English grammar
<same table shape, citing grammar-rule ID>
### Dimension 4: LaTeX format
<same table shape>
### Dimension 5: Figure quality
<same table shape>
### Banned-vocabulary and em-dash scan
<list with line references>
### Final score (1-10)
<score>
### Submission recommendation
- <Ready to submit | Needs 1-2 days more work | Needs major revision before submission>