Evidence Table

An evidence table maps each paper claim to the experiment, metric, baseline, and limitation that support it. It prevents a paper note from becoming a list of ungrounded impressions.

For your own experiments, use the same habit through a Claim evidence record.

The core relation is:

$$\text{claim}\to \text{evidence}\to \text{scope}\to \text{limit}$$

Suggested Columns

Column	Meaning
Claim	The narrow statement being evaluated.
Evidence	Figure, table, proof, benchmark, or experiment supporting it.
Task	What task the evidence actually tests.
Data	Dataset, benchmark, split, and preprocessing context.
Metric	Number used to support the claim.
Baseline	Comparison that makes the metric meaningful.
Uncertainty	Confidence interval, seed variance, or missing uncertainty.
Limitation	Why the claim may not generalize.
Wiki links	Reusable concepts updated by the claim.

Evidence Strength

A rough evidence strength score can be written as:

$$S=1_{\mathrm{metric}}+1_{\mathrm{baseline}}+1_{\mathrm{split}}+1_{\mathrm{uncertainty}}+1_{\mathrm{ablation}}$$

This is not a scientific score; it is a checklist for whether the paper gives enough information to interpret a claim.

Checks

• Is every important claim tied to a concrete result?
• Does the metric match the task?
• Is the baseline strong enough?
• Does the split support the stated generalization claim?
• Are uncertainty and failure cases reported?
• Does the result table state whether values are single-run, mean, best seed, or validation-selected?
• Is the claim phrased no broader than the evidence?

Related

• Claim extraction
• Result table reading
• Claim evidence record
• Paper note format
• Paper review workflow
• Limitation taxonomy
• Metric
• Baseline