Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when designing or writing mechanism analyses for a 《中国农村经济》 empirical manuscript. Mechanism tests grounded in household-level behavior are near-mandatory for empirical submissions. 本技能服务于《中国农村经济》(China Rural Economy, CRE)。
.claude/skills/brycewang-stanford-cre-mechanism/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 34% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 91% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 53% | 0% |
| case-16 | ✗→✓ | ▲ Improved | 53% | 0% |
《中国农村经济》的机制分析忌讳停留在宏观叙事。机制变量应落到农户 / 家庭 / 经营主体 / 村庄的可观测行为或资源配置上，例如：
本文进一步从[机制名，落到农户 / 村庄行为]角度提供证据。
首先，从理论上看，[机制变量 M] 应通过[路径]影响[因变量 Y]……
其次，实证上，本文采用[路径 A/B/C]：……，估计结果如表 X 所示。
最后，结合[已有三农文献]，本文认为该机制在中国农村情境下……。【机制数】X 个
【机制路径】A 中介 / B 替换 Y / C 子样本
【机制层面】农户 / 家庭 / 经营主体 / 村庄
【理论引用】[三农文献]
【时间一致性】通过 / 待修
【下一步】cre-heterogeneity先锁定农村问题、政策/制度场景、识别链条、机制证据和可执行含义，再判断稿件是否回应农村经济审稿人通常同时追问“三农”问题意识、政策场景、识别可信度和农村制度机制。
resources/official-source-map.md，列出仍可能改变建议的一个未核实事实。| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 37,600 | 29,980 | -20% | 1 | 1 | 0% | 4,772 | 5,049 | +6% | 0 | 0 | — |
case-02 | fail→pass | 38,880 | 33,405 | -14% | 1 | 1 | 0% | 5,316 | 6,006 | +13% | 0 | 0 | — |
case-03 | fail→fail | 28,502 | 34,460 | +21% | 1 | 1 | 0% | 4,786 | 6,108 | +28% | 0 | 0 | — |
case-08 | pass→pass | 22,183 | 23,623 | +6% | 1 | 1 | 0% | 2,831 | 4,076 | +44% | 0 | 0 | — |
case-09 | fail→fail | 30,084 | 28,289 | -6% | 1 | 1 | 0% | 3,464 | 4,800 | +39% | 0 | 0 | — |
case-04 | pass→pass | 24,298 | 29,354 | +21% | 1 | 1 | 0% | 3,727 | 5,166 | +39% | 0 | 0 | — |
case-05 | pass→pass | 29,303 | 30,750 | +5% | 1 | 1 | 0% | 3,343 | 4,957 | +48% | 0 | 0 | — |
case-06 | fail→pass | 18,474 | 20,265 | +10% | 1 | 1 | 0% | 3,116 | 4,182 | +34% | 0 | 0 | — |
case-07 | pass→pass | 28,199 | 23,333 | -17% | 1 | 1 | 0% | 3,503 | 4,316 | +23% | 0 | 0 | — |
case-10 | fail→pass | 16,662 | 29,912 | +80% | 1 | 1 | 0% | 2,174 | 4,158 | +91% | 0 | 0 | — |
case-11 | pass→pass | 27,433 | 33,214 | +21% | 1 | 1 | 0% | 3,895 | 5,717 | +47% | 0 | 0 | — |
case-12 | fail→pass | 23,904 | 28,895 | +21% | 1 | 1 | 0% | 2,799 | 4,271 | +53% | 0 | 0 | — |
case-13 | pass→pass | 25,662 | 28,869 | +12% | 1 | 1 | 0% | 2,578 | 4,205 | +63% | 0 | 0 | — |
case-22 | pass→pass | 22,833 | 30,150 | +32% | 1 | 1 | 0% | 3,720 | 5,336 | +43% | 0 | 0 | — |
case-14 | pass→pass | 20,099 | 27,484 | +37% | 1 | 1 | 0% | 2,556 | 4,272 | +67% | 0 | 0 | — |
case-15 | fail→fail | 20,731 | 16,412 | -21% | 1 | 1 | 0% | 2,059 | 3,211 | +56% | 0 | 0 | — |
case-16 | fail→pass | 19,738 | 19,337 | -2% | 1 | 1 | 0% | 2,832 | 4,319 | +53% | 0 | 0 | — |
case-17 | fail→fail | 15,815 | 14,744 | -7% | 1 | 1 | 0% | 1,654 | 3,114 | +88% | 0 | 0 | — |
case-18 | pass→pass | 20,845 | 24,323 | +17% | 1 | 1 | 0% | 3,033 | 4,738 | +56% | 0 | 0 | — |
case-19 | pass→pass | 27,514 | 22,493 | -18% | 1 | 1 | 0% | 3,614 | 4,495 | +24% | 0 | 0 | — |
case-20 | pass→pass | 18,242 | 18,719 | +3% | 1 | 1 | 0% | 2,682 | 3,836 | +43% | 0 | 0 | — |
case-21 | pass→fail | 21,960 | 27,853 | +27% | 1 | 1 | 0% | 3,575 | 5,610 | +57% | 0 | 0 | — |
case-23 | pass→fail | 22,008 | 23,205 | +5% | 1 | 1 | 0% | 2,700 | 4,354 | +61% | 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. 23 cases were attempted. The headline lift of +13 percentage points is the difference between those two pass rates over the 23 comparable cases. 2 cases got worse with the skill loaded, and they are included in that figure.
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.