Contradiction

The contradiction agent is an adversarial quality-control pass. It assumes your current conclusion may be wrong and searches for the strongest counter-evidence.

What it does

It identifies direct contradictions, boundary conditions, alternate explanations, and hidden assumptions. Findings are severity-graded so you can distinguish fatal issues from minor caveats.

Instead of asking “is this claim plausible?”, it asks:

• where does this claim fail?
• what evidence points the other way?
• what assumptions are carrying too much weight?

Output

A contradiction log mapping each challenged claim to conflicting evidence and the check needed to resolve it.

Contradiction types

The agent classifies issues into practical categories:

• Direct contradiction — source states the opposite of the claim
• Scope mismatch — claim generalizes beyond source conditions
• Alternate explanation — evidence can be explained by a different mechanism
• Fragile assumption — conclusion depends on untested prerequisites

This classification helps teams choose whether to rewrite, re-test, or remove a claim.

Severity model

Findings are typically grouped as:

• Fatal — claim is likely wrong or unsupported
• Major — claim may be directionally right but overstated
• Minor — caveats needed for precision

Severity prioritization makes contradiction output actionable for drafts, reviews, and decision memos.

When to use it

Use contradiction analysis before finalizing:

• executive summaries
• publication-facing drafts
• recommendations tied to cost or risk

It is especially valuable when multiple stakeholders already agree on a narrative, because consensus can hide blind spots.

Limitations

This agent optimizes for finding weaknesses, not polishing language. After contradiction review, use writer/verifier passes to produce final reader-ready output.

Used by

Ideal for /review, /audit, and final validation stages before publishing a brief or draft.