gcb-data-analysis — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited gcb-data-analysis (Agent Skill) and scored it 100/100 (green). The audit ran 55 deterministic rules across Security, Supply Chain, Maintenance, Transparency, and Community; it found 0 high-severity and 0 lower-severity findings. The full rule-by-rule trace and per-finding evidence are below. Free, methodology-open.
Findings & checks · 0 flagged
Every scanned point with the score it earned and what moved between them.
First recorded scan — no prior version to compare against.
The primary manifest — the file an agent reads to learn what this artifact does.
GCB reviewers are quantitatively sophisticated, and because data and code are archived publicly with a DOI (see gcb-reporting-and-data-policy), the analysis must be reproducible by a third party. Analyze as if both are true — because they are. This skill covers execution and reporting norms; design decisions live in gcb-study-design.
lme4, glmmTMB, brms, INLA)for nested, repeated-measures, and spatially/temporally autocorrelated data; do not ignore random effects or autocorrelation.
or stars; state the magnitude and its ecological/biogeochemical meaning.
and scenario uncertainty; prefer ensembles**; show measurement error where it matters.
effects, heterogeneity (I^2, tau^2), moderators pre-specified, and a publication-bias check.
where it succeeds.
pseudoreplication carrying through from design.
renv.lock, conda/requirements.txt, model version + forcing).GCB referees expect the analysis to fit the data-generating process. Use this as a routing table from question shape to the inferential machinery a quantitatively literate reviewer will look for.
| Question shape | Expected machinery | What a reviewer checks |
|---|---|---|
| Effect of a manipulated driver across randomized plots | Mixed model with plot/block random effects | Random structure matches the design; no pseudoreplication |
| Trend in a flux time series | Autocorrelation-aware regression / state-space | Residual autocorrelation modelled, not ignored |
| Spatial pattern across a gradient | Spatial random field (INLA/spaMM) | Spatial dependence handled; CRS and area stated |
| Synthesis across many studies | Random/mixed-effects meta-analysis | Effect-size choice, I^2/tau^2, bias check |
| Future projection from a process model | Multi-model ensemble | Structural + parameter + scenario spread shown |
A warming-experiment meta-analysis pools log response ratios (lnRR) of aboveground biomass from 64 studies. A defensible GCB workflow: fit a random-effects model, report the pooled lnRR back-transformed to a percentage with its interval, and quantify heterogeneity. Illustrative output — pooled lnRR 0.12, i.e. a +13% biomass response (95% CI 6–20%), I^2 = 71% with tau^2 = 0.04, and a moderator showing the effect halves in water-limited sites. The funnel plot and trim-and-fill leave the sign unchanged. The 71% heterogeneity is the result, not noise: it motivates the moisture moderator. All numbers illustrative.
split-plot structure at the true unit of inference.
pooled mean.
scenario spread rather than reporting one trajectory.
the conditions where the model fails.
【Main estimate】effect size + interval + ecological/biogeochemical meaning
【Data structure】random effects / autocorrelation handled? [Y/N]
【Uncertainty】measurement + parameter + structural + scenario partitioned?
【Model evaluation / heterogeneity】skill metrics or I^2 reported?
【Reproducible】master script + seeds + pinned versions? [Y/N]
【Next】gcb-figures-and-tables../../resources/external_tools.md — mixed-model, meta-analysis, spatial, and modelling packages../../resources/official-source-map.md — data/code archiving policy~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.