quill

Ask for an adversarial review of your paper before submission. The review steelmans your central claim, attacks each load-bearing assumption, and returns severity-ranked findings — Fatal, Major, Minor, Taste — each with the location, problem, and fix.

State the claim and request a stress-test. The review enumerates what would have to be true for the claim to hold, hunts alternative explanations and confounds, and checks whether the evidence supports the claim's strength.

Yes, grant proposals are a supported use case. The review checks feasibility argumentation, significance framing, and timeline credibility, flagging issues like missing preliminary results that would sink the proposal at panel review.

Yes, reproducibility is a first-class criterion: missing methods detail, unavailable data or code, and vague hyperparameter tuning are flagged. Statistical smells like uncorrected multiple comparisons or missing effect sizes are also flagged.

Avoid it for early brainstorming where ideas are not yet claims, or when you need copyediting rather than substantive critique. The review targets load-bearing flaws, not formatting or style polish.