Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when finalizing the main exhibits of a Quarterly Journal of Economics (QJE) manuscript — making the paper figure-forward, with clean tables and self-contained notes that read well in QJE's single-PDF, author-date format. Designs exhibits; it does not run new analysis.
.claude/skills/bilal140202-qje-tables-figures/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-13 | ✗→✓ | ▲ Improved | 23% | 0% |
| case-19 | ✗→✓ | ▲ Improved | -6% | 0% |
| case-22 | ✗→✓ | ▲ Improved | -2% | 0% |
| case-01 | ✓→✓ | = Same ✓ | -28% | 0% |
| case-02 | ✓→✓ | = Same ✓ | -35% | 0% |
QJE has moved firmly toward figure-forward presentation — the Opportunity Insights / Chetty-style QJE paper makes its central result legible in one well-designed graph (e.g., the binned mobility maps and exposure-effect plots of the QJE 2014/2018 neighborhoods papers). Identification designs are inherently visual: event-study plots, RDD discontinuity plots, and binned scatters communicate credibility better than a coefficient buried in a regression column. Tables remain essential for estimates and robustness, but the headline should often be a figure a reader grasps in five seconds. Practical QJE constraints: at initial submission everything is one PDF with figures embedded (no separate figure files), exhibits are numbered and called out in order, and in-text references are author-date (Chicago).
| Design | Headline figure | |-----------------|--------------------------------------------------------------| | DID / event std | Event-study plot: leads ≈ 0, clean post-treatment dynamics | | RDD | Discontinuity plot: binned means + local polynomial fit | | IV | First-stage and reduced-form scatter / binscatter | | RCT | Treatment-vs-control outcome distributions or effect-by-arm | | Descriptive | The new fact, plotted with the data doing the talking |
【Headline exhibit】figure type chosen + why
【Main table】column count + what moved to appendix
【Notes audit】sample / units / FE / clustering / significance present? [Y/N each]
【Magnitude legibility】units stated? [Y/N]
【Figures plotted】[event study / RDD / first stage / ...]
【Next step】qje-writing-style| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→pass | 18,292 | 7,722 | -58% | 1 | 1 | 0% | 3,110 | 2,254 | -28% | 0 | 0 | — |
case-02 | pass→pass | 22,813 | 9,474 | -58% | 1 | 1 | 0% | 3,999 | 2,593 | -35% | 0 | 0 | — |
case-03 | pass→pass | 29,255 | 6,315 | -78% | 1 | 1 | 0% | 4,809 | 2,030 | -58% | 0 | 0 | — |
case-04 | pass→pass | 13,148 | 8,585 | -35% | 1 | 1 | 0% | 2,375 | 2,569 | +8% | 0 | 0 | — |
case-05 | pass→pass | 20,704 | 21,877 | +6% | 1 | 1 | 0% | 3,768 | 5,251 | +39% | 0 | 0 | — |
case-06 | pass→pass | 13,896 | 10,722 | -23% | 1 | 1 | 0% | 2,114 | 2,390 | +13% | 0 | 0 | — |
case-07 | pass→pass | 13,676 | 7,545 | -45% | 1 | 1 | 0% | 2,088 | 2,214 | +6% | 0 | 0 | — |
case-08 | pass→pass | 13,074 | 7,381 | -44% | 1 | 1 | 0% | 1,890 | 2,127 | +13% | 0 | 0 | — |
case-09 | pass→pass | 14,483 | 10,897 | -25% | 1 | 1 | 0% | 2,320 | 2,725 | +17% | 0 | 0 | — |
case-10 | pass→pass | 12,884 | 8,893 | -31% | 1 | 1 | 0% | 1,994 | 2,433 | +22% | 0 | 0 | — |
case-11 | pass→pass | 15,382 | 7,821 | -49% | 1 | 1 | 0% | 2,283 | 2,126 | -7% | 0 | 0 | — |
case-12 | pass→pass | 11,723 | 7,401 | -37% | 1 | 1 | 0% | 2,097 | 2,240 | +7% | 0 | 0 | — |
case-13 | fail→pass | 9,688 | 6,896 | -29% | 1 | 1 | 0% | 1,855 | 2,284 | +23% | 0 | 0 | — |
case-14 | pass→pass | 11,348 | 6,195 | -45% | 1 | 1 | 0% | 2,169 | 2,084 | -4% | 0 | 0 | — |
case-15 | pass→pass | 9,606 | 6,703 | -30% | 1 | 1 | 0% | 1,660 | 1,985 | +20% | 0 | 0 | — |
case-16 | pass→pass | 9,358 | 5,467 | -42% | 1 | 1 | 0% | 1,572 | 1,744 | +11% | 0 | 0 | — |
case-17 | pass→pass | 11,095 | 7,057 | -36% | 1 | 1 | 0% | 1,713 | 2,194 | +28% | 0 | 0 | — |
case-18 | pass→pass | 14,240 | 8,854 | -38% | 1 | 1 | 0% | 2,282 | 2,461 | +8% | 0 | 0 | — |
case-19 | fail→pass | 13,766 | 8,331 | -39% | 1 | 1 | 0% | 2,634 | 2,487 | -6% | 0 | 0 | — |
case-20 | pass→pass | 9,796 | 8,159 | -17% | 1 | 1 | 0% | 1,663 | 2,329 | +40% | 0 | 0 | — |
case-21 | pass→pass | 11,700 | 11,441 | -2% | 1 | 1 | 0% | 1,872 | 2,121 | +13% | 0 | 0 | — |
case-22 | fail→pass | 10,840 | 5,912 | -45% | 1 | 1 | 0% | 2,019 | 1,970 | -2% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +14 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.