Paper Experimental Setup

Use a table if there are 4+ datasets:

\begin{table}[t]
\caption{Dataset statistics.}
\label{tab:datasets}
\centering
\small
\begin{tabular}{lccc}
\toprule
\textbf{Dataset} & \textbf{Queries} & \textbf{Corpus} & \textbf{Task} \\
\midrule
MuSiQue & 2,417 & 24K & Multi-hop QA \\
\bottomrule
\end{tabular}
\end{table}

Subsection: Baselines List each compared method with a one-sentence description and citation. Group them logically (e.g., dense retrievers, graph-based methods, iterative methods).

Subsection: Evaluation Metrics Define each metric precisely. For example:

• Passage Recall@k: percentage of queries where at least one supporting passage appears in top-k
• F1: token-level overlap between predicted and gold answers
• EM: exact match after normalization

Subsection: Implementation Details

• Models: which LLM, which embedder, specific versions
• Hardware: GPU type, memory
• Hyperparameters: learning rate, PPR damping factor, number of retrieved passages, etc.
• Runtime: how long does indexing/inference take
• Code availability: mention if releasing code

Writing Rules

• ~1 page in two-column format
• Be precise enough for reproduction — exact model names, versions, seeds
• Cite the source for every dataset and baseline method
• Use consistent terminology with the methodology section
• Do not discuss results here — only setup
• Use tables for dense information (datasets, hyperparameters)

Output

Write to sections/experimental_setup.tex. Git add, commit, push.

Writing Style — Humanize Your Output

• No AI tells. Avoid: "notably", "importantly", "it is worth noting", "leveraging", "in recent years", "a growing body of", "comprehensive", "robust", "significant", "crucial", "facilitates", "utilizing", "delve into", "shed light on", "paradigm shift"
• Vary sentence rhythm. Mix short punchy sentences with longer ones. Academic writing has cadence — do not produce uniform sentence lengths.
• Active voice. "We propose" not "A method is proposed". "The model achieves" not "It can be observed that the model achieves".
• Be direct. Cut filler. "X outperforms Y by 4.2 F1" not "It is worth noting that X demonstrates a notable improvement of 4.2 F1 points over Y".
• Sound human. Write as a researcher explaining their work to a peer at a conference — confident, precise, occasionally opinionated. Not a textbook. Not a press release.
• No bullet lists in the paper. Everything must be flowing prose (except contribution lists in the intro).
• NEVER add Co-Authored-By Claude or any AI attribution in commits or paper content.