Paper Planning

A structured approach to planning academic papers before writing begins. Covers four key activities: Story design, Experiment planning, Figure design, and Timeline management.

When to Use This Skill

If you don't yet have an idea, use the research-ideation skill first to find a problem and design a solution.

• User wants to plan a paper before writing
• User asks about structuring a paper's story or contributions
• User needs to plan experiments (comparisons, ablations)
• User wants to design pipeline figures or teaser figures
• User asks about writing timelines or submission schedules

Planning Overview

Paper planning follows four steps, ideally completed before writing begins:

Step 1: Story Design     → What is the narrative? What are the contributions?
Step 2: Experiment Plan   → What experiments prove our claims?
Step 3: Figure Design     → How do we visually communicate the method?
Step 4: Timeline          → When does each section get written?

Counterintuitive Planning First

Prioritize these counterintuitive rules before regular planning:

1. Write your rejection letter first: Draft the top-5 likely rejection comments ("limited novelty", "missing baseline", "not robust", etc.), then plan experiments that directly preempt each one.
2. Narrow claim before broad claim: Define the smallest defensible core claim first. Expand only after evidence is strong. Over-broad claims fail review more often than narrow strong claims.
3. Design ablations before polishing method text: If a module cannot be ablated cleanly, its contribution claim is weak.
4. Allocate compute to stress tests, not only benchmarks: A single convincing stress-test figure often contributes more than multiple small benchmark gains.
5. Plan a fallback narrative now: If SOTA gain is marginal, predefine a secondary value proposition (efficiency, robustness, fewer assumptions, wider applicability).

See references/counterintuitive-planning.md

---

Step 1: Story Design

The "story" is the logical narrative that connects the problem, insight, method, and results.

Reverse Engineering the Story

Work backwards to build the story:

1. What is the technical problem? — The specific challenge that existing methods cannot solve well
2. What are our contributions? — The concrete technical novelties
3. What are the benefits and new insights? — What advantages does our approach provide?
4. How do we lead into the challenge? — How to frame the task and previous methods to naturally arrive at the challenge

Then write forward: Task → Previous methods → Challenge → Our contributions → Advantages

Core Elements to Define

Before writing any section, clearly articulate:

Element	Question	Example
Task	What problem does this paper address?	"Real-time 3D scene reconstruction"
Challenge	Why can't existing methods solve it well?	"Cannot handle dynamic objects efficiently"
Insight	What key observation drives our approach?	"Motion patterns are temporally sparse"
Contribution	What do we propose?	"Sparse temporal attention for dynamic regions"
Advantage	Why is our approach better?	"Reduces computation while preserving quality"

Starting Point: Pipeline Figure Sketch

Start by drawing a pipeline figure sketch. This forces you to clarify the overall method before writing.

The pipeline figure sketch serves as the paper's visual backbone:

• Draw it before writing anything
• It reveals whether the method is clear enough to explain
• It identifies the novel modules vs. standard components
• It determines subsection structure for the Method section

See references/story-design.md

---

Step 2: Experiment Planning

Plan experiments before writing to avoid discovering gaps late.

Two Categories of Experiments

Comparison Experiments — Prove our method is better:

• Which baseline methods to compare against?
• Which datasets and metrics?
• What is the evaluation protocol?

Ablation Studies — Prove each module is effective:

• Part 1: One big table showing impact of core contributions
• Part 2: Several small tables for design choices and hyperparameters

Planning Checklist

• List all comparison baselines (recent, relevant, SOTA)
• Define evaluation metrics (standard for the task)
• Identify datasets (standard benchmarks + challenging demos)
• List ablation configurations (remove each core component)
• Plan design-choice tables (hyperparameters, input quality, alternatives)
• Plan demo scenarios (challenging data to showcase upper limit)

See references/experiment-planning.md

Experiment Plan Template

Use the template at assets/experiment-plan-template.md to organize your experiment plan.

---

Step 3: Figure Design

The pipeline figure is for highlighting novelty, not for making readers understand. The Method text is what makes readers understand.

Pipeline Figure Principles

• Highlight novelty: The pipeline figure showcases what is new, not just the workflow
• Differentiate from prior work: The figure must look different from previous methods
• Novel modules stand out: If the overall pipeline is standard, zoom in on novel modules
• Focus on clarity of the novel parts; standard components can be simplified

Teaser Figure

The teaser (usually Figure 1) shows the key result at a glance:

• Place it at the top of the first page
• Should be immediately compelling
• Reference it from the Introduction

Study Workflow Diagram

Most papers benefit from a study workflow diagram — a high-level figure showing the overall research design (data collection → preprocessing → method → evaluation). This is distinct from the pipeline figure (which shows the technical method) and often appears as Figure 1 or in the Experiments section.

The pipeline automatically generates figures/workflow.png via the study-workflow skill before assembly, using the executed plan steps as the prompt. The prompt describes two swim-lanes (Research and Writing) derived from the actual plan. No manual invocation is needed.

To manually generate or regenerate a custom workflow diagram, invoke the study-workflow skill directly: provide a plain-text description of the workflow stages and it will produce a PNG suitable for inclusion in the paper.

Visual Quality Matters

Visual polish directly influences review outcomes. See references/figure-design.md for the full visual quality guide (pipeline figures, tables, typography)

Visualization constraints (enforced at writing time):

• All charts/plots use seaborn or plotly only — no plain matplotlib default styles.
• Single-color bar charts are forbidden. Use a distinct color per category/group (e.g., seaborn.color_palette("tab10")).
• All tables must maintain consistent column count and font size across the paper. Use \resizebox{\textwidth}{!}{...} to normalize wide tables.
• Hyperparameter details: report only final values + 1-sentence justification. Omit tuning grids and search trajectories.

LaTeX Formatting Standards

The generated LaTeX skeleton includes float, microtype, url, and hyperref with breaklinks. Writing agents must follow these rules to prevent compilation errors and overflow:

• Figures: use [htbp] placement, \includegraphics[width=0.9\textwidth]{figures/<name>}, self-contained caption, \label{fig:...}
• Wide tables: wrap with \resizebox{\textwidth}{!}{...} around the tabular environment
• URLs: always use \url{...} — bare URLs cause overflow and compilation warnings
• Special characters: escape %, &, #, $, _ outside math mode
• No preamble in sections: section bodies must start with \section{...} — never include \documentclass, \usepackage, or \begin{document}
• No bibliography in sections: \bibliography, \bibliographystyle, \begin{thebibliography} are handled by the skeleton

---

Step 4: Timeline

4-Week Countdown

Start writing at least 1 month before the deadline.

Week	Tasks
4 weeks before	1. Organize story (core contribution, module motivations). 2. List comparison experiments and ablation studies. 3. Write Introduction first draft.
3 weeks before	1. Finalize the pipeline figure sketch. 2. Write Method first draft (use \todo{} for unsettled details). Deadline: give Introduction + Method draft to advisor.
2 weeks before	Write first drafts of Experiments, Abstract, Related Work.
Last week	Revise paper, polish pipeline figure and teaser, run demos.

Critical: By the end of Week 3, you must send the Introduction and Method drafts to your advisor — otherwise the advisor likely will not have enough time to finish reviewing the paper.

See references/timeline-4week.md for the detailed schedule and progress tracking template.

---

Handoff to Writing

When planning is complete, pass these artifacts to paper-writing:

Artifact	Source Step	Used By
Story summary (task → challenge → insight → contribution → advantage)	Step 1	Introduction
Module Motivation Mapping table	Step 1	Method subsections
Experiment plan (comparisons + ablations + demos)	Step 2	Experiments section
Pipeline figure sketch	Step 1 / Step 3	Method overview + Figure 2
Claim-to-experiment mapping	Step 2	Abstract, Introduction, Experiments
Fallback narrative (if planned)	Counterintuitive Rule 5	Introduction / Conclusion pivot
Rejection-risk table	Counterintuitive Rule 1	Self-review prioritization

---

Reference Navigation

Topic	Reference File	When to Use
Story design	story-design.md	Starting a new paper
Experiment planning	experiment-planning.md	Before running experiments
Timeline	timeline-4week.md	Setting up a writing schedule
Figure design	figure-design.md	Designing pipeline/teaser figures
Experiment plan template	experiment-plan-template.md	Creating a structured experiment plan
Counterintuitive strategy	counterintuitive-planning.md	Increasing acceptance odds with non-obvious planning choices
Study workflow diagram	study-workflow/SKILL.md	Generating a polished workflow diagram image via gpt-5.4-image-2

Handoff to Presentation

If preparing a conference talk or slide deck, the academic-slides skill guides slide creation from your planning artifacts — including translating your story design and pipeline figure into presentation structure.