Use when revising an AISTATS paper for concise AI-statistics framing, theorem-and-experiment clarity, 8-page two-column compression, double-blind wording, reproducibility clarity, statistically careful claims, and assumption-labeling discipline that survives statistician reviewers.
SkillScore86Maintained
Automatic
◆Writing & Editing pack
Last verified 8/15/2026
What it changes
Passed 19 of 22 test cases with the skill, and 15 of 22 without it.
Without skill68%
With skill86%
Checks it passed that the model alone missed
Replaces vague performance claims with explicit mean error metrics and standard errors across seeded runs
Replaces 'easy to see' with an explicit mathematical justification referencing specific lemmas or inequalities
States that the first page must articulate the problem, gap, method or theorem, evidence, and why existing methods fail
and 4 more in the full report
6 fixed13 passed either way1 still failing2 broke
Pass-rate lift
+18 pp
Tokens
+7%
Agent turns
0%
Response time
-25%
Model gemini-3.6-flashMeasured 8/15/2026v119 of 22 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 6 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 6
Prompt
Review this sentence from my statistical learning paper draft: 'Our algorithm significantly outperforms all existing baseline algorithms on real-world datasets.' I want to fix vague language for a rigorous statistics conference submission. Provide a revised version that quantifies performance rigor.
What was checked, without and with the skill
Checked
Without skill
With skill
Replaces vague performance claims with explicit mean error metrics and standard errors across seeded runs
Missed
Passed
Scopes empirical claims specifically to the regimes tested rather than asserting universal dominance
Passed
Passed
case-06 · gemini-3.6-flash · 8/15/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md59 linessingle file
AISTATS Writing Style
Use this when revising the main paper. AISTATS papers need compact statements of why a
statistical or ML contribution matters and enough detail for technical validation.
Revision rules
Put the AI/statistics contribution in the first page: problem, gap, method or theorem,
evidence, and why existing methods are insufficient.
Make assumptions explicit. AISTATS reviewers are sensitive to hidden distributional,
asymptotic, optimization, or data-generating assumptions.
Pair every major claim with either proof structure, simulation design, empirical table, or
reproducibility detail.
Use the 8-page body for core logic; move long derivations, extra simulations, and extended
ablations to the appendix without making the main paper unintelligible.
Avoid overclaiming empirical wins when differences are small or variance is unreported.
Maintain double-blind style in self-citations, prior code references, acknowledgements,
funding, and artifact descriptions.
Theorem-presentation discipline
Number assumptions and cite them by label inside every theorem statement; AISTATS readers
audit assumption flows the way software reviewers audit imports.
Give each major theorem a proof sketch in the body; the appendix proof supports the sketch
but never replaces it.
Keep problem-dependent constants visible or explicitly deferred; silently absorbing
dimension or condition numbers into O-notation is a standing statistician complaint.
Define notation once in one location — two-column pages punish redundant redefinition.
Phrase remarks after theorems as interpretation, not as extra unproved claims; AISTATS
reviewers treat every declarative sentence near a theorem as something they may verify.
When a result is conjectured or only empirically supported, label it so; mixing proved and
observed statements in one paragraph is a credibility leak at this venue.
Sentence-level rewrites
Draft pattern
AISTATS-safe rewrite
"Our method significantly outperforms..."
"reduces mean error by X (SE Y) over Z seeded runs"