aw-write-methodology — Methodology Section Writer

Purpose

Write the Methodology section of an academic paper (Phase 3 of IMRAD) by generating independent paragraph files that are later merged into a complete chapter by aw-execute.

This skill is a section-writing subagent called by the Wave Executor during Phase 2. It receives a specific task (e.g., "Write 3.2 Architecture Components") and outputs a single paragraph .tex file.

When to Trigger

• aw-execute wave executor calls this skill with a specific task
• User runs /aw-write-methodology directly
• Orchestrator delegates during GSDAW pipeline execution

Inputs

Input	Source	Description
Task description	Wave executor (objective field)	Which paragraph to write (e.g., "3.2 Architecture Components")
research-brief.json	.planning/research-brief.json	Author intent, novelty claims, research question
methodology.md	.planning/methodology.md	Full technical methodology with pipeline, architecture, loss, hyperparameters
literature.md	.planning/literature.md	Related work context (for positioning against baselines)
Elsevier template	templates/elsevier/	LaTeX format reference (document class, packages, citation style)

Outputs

Paragraph files written to sections/methodology/ with naming convention {task-id}.tex:

File	Task ID	Content
sections/methodology/3-1-overview.tex	3.1	Method overview (2-3 paragraphs)
sections/methodology/3-2-architecture.tex	3.2	Architecture components with architecture table
sections/methodology/3-3-innovations.tex	3.3	Key innovations and technical contributions
sections/methodology/3-4-experiment-preview.tex	3.4	Experimental setup preview (datasets, metrics preview)

Output File Format

Each paragraph file uses Elsevier LaTeX conventions:

paragraph{Method Overview}
\label{sec:method:overview}

% Write 2-3 paragraphs here describing the high-level approach.
% Use \cite{key} for citations, \ref{fig:xxx} for figures.

paragraph{Architecture Components}
\label{sec:method:arch}

% Describe each architectural component in technical detail.
% Include inline formulas with \[
% Reference the architecture table (Table~\ref{tab:arch}).

Architecture Table Format

Include an architecture table for the U-Net encoder-decoder:

\begin{table}[htbp]
\centering
\caption{U-Net Encoder-Decoder Architecture}
\label{tab:arch}
\begin{tabular}{lccc}
    \toprule
    Stage & Layer & Output Channels & Spatial Resolution \\
    \midrule
    Input & — & 1 & L = 512 \\
    Enc1 & Conv(3x3)+BN+ReLU $\times$2 & 32 & L \\
    Enc2 & MaxPool1D(2)+Conv$\times$2 & 64 & L/2 \\
    Enc3 & MaxPool1D(2)+Conv$\times$2 & 128 & L/4 \\
    Enc4 & MaxPool1D(2)+Conv$\times$2 & 256 & L/8 \\
    Bottleneck & MaxPool1D(2)+Conv$\times$2+Dropout(0.1) & 512 & L/16 \\
    Dec4 & UpConv+Concat+Conv$\times$2 & 256 & L/8 \\
    Dec3 & UpConv+Concat+Conv$\times$2 & 128 & L/4 \\
    Dec2 & UpConv+Concat+Conv$\times$2 & 64 & L/2 \\
    Dec1 & UpConv+Concat+Conv$\times$2 & 32 & L \\
    Output & Conv(1x1) & 1 & L \\
    \bottomrule
\end{tabular}
\end{table}

Loss Function Formula

Include the mixed loss formula:

\subsection{Loss Function}

The network is trained with a mixed loss function combining MSE, CCC, and SSIM terms:

\[
\mathcal{L}_{total} = \alpha \cdot \mathcal{L}_{MSE} + \beta \cdot \mathcal{L}_{CCC} + \gamma \cdot \mathcal{L}_{SSIM}
\]
where $\alpha = 0.5$, $\beta = 0.3$, $\gamma = 0.2$.

The CCC term measures waveform morphology preservation:
\[
\mathcal{L}_{CCC} = 1 - \frac{2\text{Cov}(y_{true}, y_{pred})}{\text{Var}(y_{true}) + \text{Var}(y_{pred}) + (\bar{y}_{true} - \bar{y}_{pred})^2}
\]

The SSIM term measures local structure preservation:
\[
\mathcal{L}_{SSIM} = 1 - \frac{(2\mu_x\mu_y + C_1)(2\sigma_{xy} + C_2)}{(\mu_x^2 + \mu_y^2 + C_1)(\sigma_x^2 + \sigma_y^2 + C_2)}
\]

Key Hyperparameters Table

Include hyperparameters table:

\begin{table}[htbp]
\centering
\caption{Training Hyperparameters}
\label{tab:hyperparams}
\begin{tabular}{ll}
    \toprule
    Parameter & Value \\
    \midrule
    Optimizer & Adam \\
    Learning rate & $1\times10^{-3}$ \\
    Batch size & 64 \\
    Window size & 512 samples \\
    Overlap & 256 samples (50\%) \\
    SNR range & -10 dB to +20 dB \\
    Early stopping & Patience = 10 epochs \\
    Weight init & He normal (fan-out) \\
    Dropout & 0.1 at bottleneck \\
    Gradient clipping & max\_norm = 1.0 \\
    \bottomrule
\end{tabular}
\end{table}

Workflow

aw-write-methodology invoked with task "3.2 Architecture Components"
    │
    ▼
Read: .planning/methodology.md (full document)
    │
    ▼
Read: .planning/research-brief.json (novelty, assumptions)
    │
    ▼
Read: templates/elsevier/sections/methodology.tex (LaTeX format)
    │
    ▼
Write: sections/methodology/3-2-architecture.tex
    │
    ▼
Return completion with word count and first 200 chars preview

Step-by-Step Execution

Step 1: Read Inputs

Read .planning/methodology.md and extract the relevant subsection for the assigned task:

• For task 3.1: read Section 1 (Technical Pipeline) + Section 4 (Method Overview from Expected Results)
• For task 3.2: read Section 2 (Network Architecture) — includes architecture table, loss function, skip connections
• For task 3.3: read Section 2.3 (Key architectural features) + Section 3.4 (Ablation studies — innovations)
• For task 3.4: read Section 3 (Experiment Design) — datasets table, baseline summary, metrics preview

Also read .planning/research-brief.json to understand the novelty claims that need emphasizing.

Step 2: Extract Content for Task

For each paragraph file, extract relevant content from methodology.md:

3-1-overview.tex (Method Overview):

• Describe the problem: laser ultrasound NDT denoising
• High-level approach: U-Net based signal denoising
• Key motivation: AWGN noise model, sim-to-real gap challenge
• Expected generalization behavior

3-2-architecture.tex (Architecture):

• Full encoder-decoder table (Table~\ref{tab:arch})
• Skip connections with residual concatenation
• Attention gating (Ablation A6)
• Loss function with mixed terms
• Hyperparameters table (Table~\ref{tab:hyperparams})

3-3-innovations.tex (Key Innovations):

• Mixed loss with CCC + SSIM for waveform morphology preservation
• Attention-gated skip connections for suppressing irrelevant signal regions
• SNR-robust training with uniform distribution (-10 to +20 dB)
• Sim-to-experimental domain adaptation via diverse noise modeling

3-4-experiment-preview.tex (Experiment Setup Preview):

• Dataset overview (training: 10K pairs, validation: 1K, test sets: sim+exp)
• Materials: Aluminum 2024-T3 (60%), CFRP laminate (40%)
• Baseline methods preview (Wiener, DWT, BM3D, Sparse coding, Bandpass)
• Evaluation metrics preview (SNR improvement, CCC, WSI, F1, POD)

Step 3: Write Paragraph File

Write the .tex paragraph file with:

1. \paragraph{Section Title} heading with label
2. 2-4 paragraphs of running text (academic register, past tense for methods)
3. Inline math with \[ ... \] for formulas
4. \ref{tab:} and \ref{fig:} for cross-references
5. \cite{key} for citations
6. End with a blank line before potential next section

Step 4: Verify Output

Check that the written file:

• Compiles (valid LaTeX syntax)
• Contains at least 150 words (target: 200-400)
• Has no TODO/FIXME placeholders
• Uses \ref{} correctly (not hardcoded numbers)
• Uses Elsevier citation style (\cite{key}, not \citep)

Step 5: Report Completion

Return to wave executor:

Paragraph 3.2 (Architecture) written.
Word count: 387
File: sections/methodology/3-2-architecture.tex
Preview: "The encoder-decoder architecture follows a U-Net topology..."

Elsevier LaTeX Conventions

Document Class

\documentclass[review]{elsarticle}

Packages Used

\usepackage{amsmath,amsfonts,amssymb}  % Math
\usepackage{graphicx}                   % Figures
\usepackage{float}                      % Float positioning
\usepackage{booktabs}                   % Tables (requires: \toprule, \midrule, \bottomrule)
\usepackage{cite}                       % Citations

Citation Style

• Use \cite{key} → numbered citation [1]
• Use \citealp{key1,key2} for grouped without parentheses
• DO NOT use author-date format

Table Format (booktabs)

\begin{table}[htbp]
\centering
\caption{Caption Here}
\label{tab:label}
\begin{tabular}{lccc}
    \toprule
    Header 1 & Header 2 & Header 3 \\
    \midrule
    Row 1 col 1 & Row 1 col 2 & Row 1 col 3 \\
    Row 2 col 1 & Row 2 col 2 & Row 2 col 3 \\
    \bottomrule
\end{tabular}
\end{table}

Cross-References

• Figures: \ref{fig:arch} → "Fig. \ref{fig:arch}"
• Tables: \ref{tab:arch} → "Table \ref{tab:arch}"
• Equations: \ref{eq:loss} → "Eq. \ref{eq:loss}"
• Sections: \ref{sec:method:overview} → "Section \ref{sec:method:overview}"

Error Handling

Missing Input Files

If .planning/methodology.md does not exist:

错误：未找到方法论设计文件 (.planning/methodology.md)。

请先运行 /aw-methodology 生成方法论设计文件，
或确认研究简报已通过 Discuss #1 审批。

If .planning/research-brief.json does not exist:

警告：未找到研究简报 (research-brief.json)。
方法论写作将仅基于 methodology.md 内容进行。

Incomplete Content

If the assigned task references content not present in methodology.md:

• Write what is available with a note: "% TODO: detail missing — verify during review"
• Report the gap in the completion message

Compilation Check

After writing, scan for common LaTeX errors:

• Unmatched braces {}
• Missing $ for inline math
• Undefined labels (will be resolved during merge)
• \begin{} without matching \end{}

File Locations

manuscripts/[paper-name]/
├── .planning/
│   ├── research-brief.json    ← Input
│   ├── methodology.md         ← Input
│   └── literature.md          ← Input (for related work positioning)
├── templates/elsevier/        ← Format reference
│   └── sections/methodology.tex
└── sections/
    └── methodology/
        ├── 3-1-overview.tex       ← Output (paragraph file)
        ├── 3-2-architecture.tex   ← Output (paragraph file)
        ├── 3-3-innovations.tex     ← Output (paragraph file)
        └── 3-4-experiment-preview.tex  ← Output (paragraph file)

Integration Points

Connection	Agent/File	Direction
Called by	aw-execute (Wave Executor)	Input: task assignment
Feeds into	aw-execute (Phase Merger)	Output: paragraph files
Reads	.planning/methodology.md	Input
Reads	.planning/research-brief.json	Input
Format reference	templates/elsevier/	Input
Later review	aw-review	After wave completion

Quality Gate Checklist

Before reporting completion, verify:

• At least 150 words in paragraph
• Academic register (past tense for methods, no informal language)
• Elsevier citation format (\cite{key})
• Architecture table uses booktabs format
• Loss function formula uses \[ ... \] display math
• No hardcoded figure/table numbers (use \ref{})
• No TODO/FIXME placeholders
• File saved to correct path sections/methodology/{filename}.tex