Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when finalizing exhibits for a 《中国行政管理》 manuscript — regression tables, survey measurement tables, qualitative coding tables, and process / mechanism diagrams. Enforces three-line table style, coding-table conventions, and figure discipline. 本技能服务于《中国行政管理》(Chinese Public Administration, CPA)。
.claude/skills/brycewang-stanford-cpa-tables-figures/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | 36% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 34% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 14% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 54% | 0% |
表 3 数字化考核对基层形式主义负担的影响
═══════════════════════════════════════════
(1) (2) (3)
变量 负担 负担 负担
───────────────────────────────────────────
数字化考核 0.118*** 0.094*** 0.087***
(0.031) (0.029) (0.030)
......
───────────────────────────────────────────
控制变量 否 是 是
地区固定效应 是 是 是
年份固定效应 是 是 是
观测值 2,431 2,431 2,431
R² 0.184 0.226 0.231
═══════════════════════════════════════════
注：括号内为聚类于地级市层面的稳健标准误；***、**、* 分别表示 1%、5%、10% 显著水平。*** ** * 统一三档；标准误在系数下方括号内，不要用 t / z 值定量主表之前必须有变量定义表（问卷研究还需测量条目与信效度）：
| 变量类型 | 变量名称 | 变量定义 / 测量 | 数据来源 | |----------|----------|----------------|----------| | 被解释变量 | burden | 基层形式主义负担（5 题李克特量表，α=0.86） | 问卷 | | 核心解释变量 | digital_kpi | 数字化考核强度（政务 App 数量对数） | 政务平台 | | 控制变量 | fiscal | 地方财政自给率 | 财政决算 |
约束：
案例 / 扎根研究必须有两张表：
| 资料编号 | 资料类型 | 来源 / 对象 | 数量 / 时长 | 用途 | |---------|---------|------------|------------|------| | D-01 | 政策文件 | XX 市实施方案 | 12 份 | 制度背景、政策过程 | | I-03 | 半结构访谈 | 街道办主任 | 5 人次 / 6.5 h | 执行逻辑 | | O-02 | 实地观察 | 政务大厅 | 3 次 | 服务互动 |
| 一级编码（选择性） | 二级编码（主轴） | 三级编码（开放） | 典型原始语句（编号） | |------------------|----------------|----------------|--------------------| | 注意力挤占 | 中心工作优先 | "App 打卡占用入户时间" | I-03、I-05 |
表格图形从拟合结果生成，不要手抄数字。完整映射见 execution-with-mcp。《中国行政管理》是公共管理刊，实证用观察性与(准)实验设计；识别 + 聚类/多层推断，定性工作另循其标准。
etable（多列）或 did_summary_to_latex 直接从 result_id 生成。plot_from_result / enhanced_event_study_plot / event_study_table，坐标单位与标准误/聚类注记自带。
完整“拟合结果 → 图表”链见 JF 执行 walkthrough。
【定量主表列数】X 列（如适用）
【三线表合规】是 / 否
【测量 / 信效度表】有 / 无 / 不适用
【定性资料表 + 编码表】有 / 无 / 不适用
【机制 / 框架图】有 / 无
【图形 DPI】X
【缺漏注释】[...]
【下一步】cpa-policy-implication| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 38,599 | 39,124 | +1% | 1 | 1 | 0% | 4,741 | 6,464 | +36% | 0 | 0 | — |
case-02 | fail→pass | 33,920 | 31,358 | -8% | 1 | 1 | 0% | 4,494 | 5,074 | +13% | 0 | 0 | — |
case-03 | fail→pass | 33,664 | 30,649 | -9% | 1 | 1 | 0% | 4,272 | 5,704 | +34% | 0 | 0 | — |
case-04 | fail→pass | 124,612 | 25,441 | -80% | 1 | 1 | 0% | 3,964 | 4,510 | +14% | 0 | 0 | — |
case-05 | fail→pass | 25,990 | 16,418 | -37% | 1 | 1 | 0% | 2,866 | 4,404 | +54% | 0 | 0 | — |
case-06 | pass→pass | 24,441 | 22,870 | -6% | 1 | 1 | 0% | 2,596 | 4,221 | +63% | 0 | 0 | — |
case-07 | pass→pass | 26,043 | 20,026 | -23% | 1 | 1 | 0% | 2,707 | 4,098 | +51% | 0 | 0 | — |
case-08 | fail→pass | 36,112 | 26,601 | -26% | 1 | 1 | 0% | 4,495 | 4,910 | +9% | 0 | 0 | — |
case-09 | fail→pass | 30,657 | 20,703 | -32% | 1 | 1 | 0% | 3,417 | 4,009 | +17% | 0 | 0 | — |
case-10 | pass→pass | 27,354 | 28,335 | +4% | 1 | 1 | 0% | 3,411 | 5,082 | +49% | 0 | 0 | — |
case-11 | pass→pass | 20,761 | 20,688 | -0% | 1 | 1 | 0% | 2,919 | 3,798 | +30% | 0 | 0 | — |
case-12 | fail→pass | 27,020 | 30,630 | +13% | 1 | 1 | 0% | 2,982 | 5,525 | +85% | 0 | 0 | — |
case-13 | pass→pass | 30,699 | 20,878 | -32% | 1 | 1 | 0% | 3,461 | 3,844 | +11% | 0 | 0 | — |
case-14 | fail→pass | 28,351 | 29,835 | +5% | 1 | 1 | 0% | 3,392 | 4,946 | +46% | 0 | 0 | — |
case-15 | pass→pass | 34,815 | 26,114 | -25% | 1 | 1 | 0% | 3,760 | 4,591 | +22% | 0 | 0 | — |
case-16 | fail→fail | 28,041 | 25,230 | -10% | 1 | 1 | 0% | 3,404 | 5,017 | +47% | 0 | 0 | — |
case-17 | fail→pass | 29,632 | 22,838 | -23% | 1 | 1 | 0% | 3,639 | 4,754 | +31% | 0 | 0 | — |
case-18 | fail→pass | 31,890 | 36,838 | +16% | 1 | 1 | 0% | 3,624 | 6,227 | +72% | 0 | 0 | — |
case-19 | fail→pass | 27,949 | 12,970 | -54% | 1 | 1 | 0% | 2,887 | 2,979 | +3% | 0 | 0 | — |
case-20 | pass→pass | 20,674 | 21,578 | +4% | 1 | 1 | 0% | 2,102 | 3,586 | +71% | 0 | 0 | — |
case-21 | pass→pass | 32,651 | 35,575 | +9% | 1 | 1 | 0% | 3,649 | 6,071 | +66% | 0 | 0 | — |
case-22 | pass→pass | 25,298 | 30,615 | +21% | 1 | 1 | 0% | 3,321 | 5,183 | +56% | 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 +55 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is 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.