Agent Skills
› brycewang-stanford/Awesome-Journal-Skills
› epsl-data-analysis
epsl-data-analysis
GitHub专为EPSL期刊设计的数据分析技能，指导同位素与年代学数据的归约、不确定性预算构建及统计模型（如MSWD、贝叶斯年龄模型）的应用。强调结果的可追溯性、误差传递规范及报告透明度，确保符合严格的审稿标准。
Trigger Scenarios
归约同位素、年代学或地球物理数据以生成可报告结果
决定如何传播和引用各项不确定度分量
计算加权平均、等时线或贝叶斯年龄-深度模型
回复审稿人关于误差预算、MSWD或分辨率测试的要求
Install
npx skills add brycewang-stanford/Awesome-Journal-Skills --skill epsl-data-analysis -g -y
SKILL.md
Frontmatter
{
"name": "epsl-data-analysis",
"description": "Use when reducing and reporting data for an Earth and Planetary Science Letters (EPSL) manuscript — full analytical-uncertainty budgets for isotope and geochronology data, defensible statistics (MSWD, weighted means, Bayesian age models), and inversion\/model diagnostics. It guides reduction and reporting norms; it does not fabricate results."
}
Data Analysis (epsl-data-analysis)
EPSL's methods-transparency culture is strictest exactly where the journal is strongest: isotope
geochemistry and geochronology. Reviewers expect every headline number to arrive with its complete
error budget — internal precision, external reproducibility, standard and blank corrections, and
systematic terms like decay-constant uncertainty — and they expect model results to arrive with their
resolution limits. Design choices live in epsl-study-design; deposition and supplements in
epsl-reporting-and-reproducibility.
When to trigger
- Reducing isotope, geochronologic, elemental, or geophysical data to reportable results
- Deciding which uncertainty terms to propagate and how to quote them
- Computing weighted means, isochrons, concordia intercepts, or Bayesian age–depth models
- A reviewer asked for the error budget, the MSWD, or a resolution test
Reporting norms EPSL expects
- State the full uncertainty ladder. Quote internal (within-run) precision, external reproducibility from repeat standards/replicates, and — when comparing across methods or to other studies — systematic terms (decay constants, tracer calibration, spectrometer bias). Say which ladder rung each quoted ±X σ includes, and whether σ means 1σ, 2σ, or 95% CI.
- Anchor to community values. Name the reference materials with the values used, the normalization scheme (e.g., what δ is measured against), and the decay constants adopted, with citations — so the numbers remain comparable when calibrations shift.
- Statistics that match geochemical data. Weighted means with MSWD reported and interpreted (MSWD ≫ 1 means scatter beyond analytical error — say why); isochrons with the regression model named; mixtures and outliers handled by a stated rule, not silent deletion.
- Age models with honest priors. Bayesian age–depth or eruption-tempo models report priors, convergence, and sensitivity to them; deposit the model input/config.
- Inversions come with diagnostics. Resolution/recovery tests, misfit and trade-off curves, and an explicit statement of what the data cannot see; never present a smoothed model as if it were data.
- Blanks and detection. Report total procedural blanks and their variability; state how blank-sensitive results (small samples, young ages, low abundances) respond to the blank range.
Uncertainty-reporting table reviewers work down
| Element | What to report | If omitted, the reviewer assumes |
|---|---|---|
| σ-level and type | 1σ/2σ/95% CI; internal vs external | numbers are not comparable |
| Reference materials | measured vs accepted values, per session | accuracy unverified |
| Blanks | magnitude + variability + correction | small samples untrustworthy |
| Decay constants / tracer | which values, cited | ages not portable across studies |
| MSWD / goodness of fit | value + interpretation | scatter hidden in the mean |
| Model resolution | recovery tests, trade-offs | structure may be artifact |
Worked micro-example (illustrative — a weighted-mean age that survives review)
Twelve single-zircon U-Pb analyses from one ash bed (illustrative numbers):
- Ten young analyses give a weighted mean of 66.021 ± 0.024 Ma (2σ, analytical only), MSWD = 1.3 — scatter consistent with analytical error, so a single population is defensible.
- Two older grains are excluded as inherited by a pre-stated criterion (resolvably older than the main cluster), and shown in the figure anyway.
- Reported as: "66.021 ± 0.024/0.031/0.075 Ma (2σ: analytical / +tracer / +decay constants), MSWD 1.3, n = 10/12" — the three-tier bracket lets a reader compare against Ar-Ar or astrochronologic ages without emailing the authors.
The habit that prevents most queries: every mean is accompanied by its MSWD, its n-of-N, and a plot showing the excluded analyses.
Referee-pushback patterns and the venue-specific fix
- "Uncertainties are quoted but not defined." → State σ-level, and split analytical vs systematic terms wherever the result is compared to external ages or data.
- "MSWD indicates overdispersion." → Do not just widen errors; identify the geological or analytical source of scatter and let the interpretation absorb it.
- "The anomaly is at the edge of resolution." → Show the recovery test at that node; soften or drop claims the test cannot support.
Anti-patterns
- A headline age or rate with a bare ± and no statement of what it includes
- Standards measured but never reported against accepted values
- Outlier analyses removed without a stated rule or a visible plot
- MSWD omitted, or quoted without interpretation
- Tomographic/model features discussed where resolution tests show smearing
- Comparing your ages to literature ages without harmonizing decay constants
Output format
【Headline number】value ± (σ-level; analytical/+systematic) + units + n
【Traceability】standards vs accepted values; blanks; constants cited
【Statistics】weighted mean/isochron/Bayesian model + MSWD/diagnostics
【Exclusions】rule stated + shown in figures? [Y/N]
【Model diagnostics】resolution/sensitivity reported? [Y/N or N/A]
【Next】epsl-figures-and-tables
Supplementary resources
../../resources/external_tools.md— IsoplotR, ET_Redux, iolite, ObsPy, inversion codes../../resources/official-source-map.md— methods-transparency and data expectations
Version History
- 9f86f09 Current 2026-07-19 15:17
