Claim Extraction

Claim extraction turns a paper from a narrative into a set of checkable statements. A paper note should not only say what the paper is about; it should identify what claims are supported, under what protocol, and with what limits.

A useful claim has the form:

$$\text{method }A\text{ improves metric }M\text{ on benchmark }B\text{ under protocol }P\text{ compared with baseline }C$$

Claim Types

• Method claim: a model, objective, architecture, or data construction is new or useful.
• Empirical claim: a measured result improves over a baseline.
• Mechanism claim: an ablation explains why the method works.
• Generalization claim: the method transfers to new data, tasks, domains, scaffolds, families, or systems.
• Efficiency claim: the method improves memory, latency, throughput, cost, or implementation complexity.
• Scientific claim: the result supports a biological, chemical, or structural interpretation.

Extraction Steps

1. Identify the paper’s central question.
2. List the main claims in one sentence each.
3. Attach each claim to a metric, benchmark, split, baseline, and uncertainty estimate when available.
4. Mark claims that depend on ablations, assumptions, or unverified interpretations.
5. Map the claim into Evidence table.
6. Move reusable definitions into Concepts.

Checks

• Is the claim directly supported by a table, figure, experiment, or proof?
• Is the comparison baseline strong and fair?
• Does the split support the generalization claim?
• Is the effect size larger than uncertainty or seed variance?
• Does the paper claim causality from an ablation that is too weak?
• Is the claim still true if phrased narrowly?

Related

• Paper note format
• Paper review workflow
• Evidence table
• Limitation taxonomy
• Paper comparison matrix
• Hypothesis
• Baseline
• Confidence interval
• Statistical significance
• Effect size
• Paired comparison
• Multiple comparisons