Use when building tables and figures for an American Journal of Political Science (AJPS) manuscript. AJPS counts table/figure headers and notes toward the word limit, and the deposited code must regenerate every exhibit exactly for third-party verification. Designs self-contained, reproducible exhibits; it does not run the analysis.
SkillScore83Maintained
Automatic
◆Research pack
Last verified 8/15/2026
What it changes
Passed 19 of 23 test cases with the skill, and 9 of 23 without it.
Without skill39%
With skill83%
Checks it passed that the model alone missed
Output includes [Y/N] indicators
Output contains the tag 【Main exhibit】
Output recommends ajps-writing-style as the next step
and 15 more in the full report
11 fixed8 passed either way3 still failing1 broke
Pass-rate lift
+43 pp
Tokens
+42%
Agent turns
0%
Response time
-25%
Model gemini-3.6-flashMeasured 8/15/2026v119 of 23 cases passed with skill
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 11 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 11
Prompt
I am preparing my online appendix for an initial submission to a political science journal. My current appendix document is 32 pages long, including balance tables and multi-model robustness checks. What is the maximum allowed page count for the online Supporting Information for an original submission?
What was checked, without and with the skill
Checked
Without skill
With skill
Output states that the SI limit for original submissions is 25 pages or fewer
Missed
Passed
Output recommends trimming the appendix to fit within the page cap
Missed
Passed
case-09 · 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.md103 linessingle file
Tables & Figures (ajps-tables-figures)
Exhibits are where an expert reviewer checks whether the result is real — and at AJPS they carry two
extra constraints. First, the headers and notes of tables and figures count toward the word limit
(the body, notes, and printed captions are all in scope; references and online SI are not). Second,
every exhibit must be regenerated by the deposited code, because a third-party verifier re-runs it
against the main-text numbers. So each exhibit must earn its words and its reproducibility.
When to trigger
Designing the main results table/figure or a key descriptive exhibit
Deciding what belongs in the article vs. the online Supporting Information (SI, <= 25 pages for original submissions)
A reviewer found an exhibit unclear, mislabeled, or not self-contained
Trimming to the word cap by moving exhibits to the SI
Principles
Self-contained. A reader understands each exhibit from its title, axis/column labels, and note
alone — state units, sample, N, and what the estimate is. (Keep notes tight: they count as words.)
Figures over dense tables for effects. Coefficient/forest plots, marginal-effects and
predicted-probability plots, event-study and RD plots show magnitude and uncertainty better than a
wall of coefficients. Show intervals.
Accessible. Colorblind-safe palettes; legible in grayscale; no chartjunk, no 3D. Reviewers must
parse it fast.
Main text vs. SI. Keep the few exhibits that carry the argument in the article; move balance
tables, full specifications, and robustness grids to the Supporting Information (<= 25 pages for original submissions,
uploaded as "Appendix"). This is also how you fit under 10,000 / 4,000 words.
Reproducible and labeled to the code. Each exhibit is generated by the master script; numbers
match the deposited package exactly (the verifier re-runs it — see
ajps-replication-and-verification). Name code by exhibit ("Files to Reproduce Table 1").
Discipline-specific exhibits
Maps for geographic/spatial variation; network diagrams where structure is the point.
Use when building tables and figures for an American Sociological Review (ASR) manuscript. ASR excludes tables and figures from the 15,000-word count but still expects them to be clear, self-contained, and ASA-formatted. Designs exhibits; it does not run the analysis.