first commit

文档润色流和知识库构建流/claude-scholar/skills/results-analysis/references/statistical-reporting.md

@@ -0,0 +1,42 @@
+# Statistical Reporting Standard
+
+## Minimum reporting package
+
+For every major comparison, report:
+- metric definition and direction,
+- unit of analysis,
+- sample size / run count,
+- descriptive statistics,
+- uncertainty estimate,
+- inferential test,
+- effect size,
+- correction strategy when multiple contrasts exist,
+- limitation if assumptions or sample size are weak.
+
+## Required fields
+
+### Descriptive
+- `mean ± std` when repeated runs are comparable
+- `95% CI` when inference is discussed
+- median / IQR when distribution is strongly non-normal
+
+### Inferential
+- exact test name
+- test statistic and degrees of freedom when applicable
+- p-value format
+- effect size
+- correction method for multiple comparisons
+
+## Do not do these
+- report only best run
+- report only p-values
+- hide non-significant comparisons
+- treat unstable trends as conclusions
+- switch tests without stating why
+
+## Default wording rule
+
+Use three layers:
+1. **Observation** — what changed numerically
+2. **Support** — what the test/effect size says
+3. **Boundary** — what remains uncertain