gcb-study-design — independently scanned and version-tracked by SaferSkills.
SaferSkills independently audited gcb-study-design (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 experts in ecology, biogeochemistry, and ecosystem/Earth-system modelling. They will probe whether the design can actually support a driver → biological-response claim at the stated scale. This skill covers design choices and their tradeoffs; analysis lives in gcb-data-analysis.
reciprocal transplants). Report dose, duration, replication, and the realism gap versus real-world change; avoid pseudoreplication (treatment confounded with plot/chamber).
LTER/NEON time series). State confounders and the limits of space-for-time substitution; use design or covariates to address them.
forcing, spin-up, parameterization, and evaluation against observations; prefer ensembles and report structural vs parameter vs scenario uncertainty**.
criteria, effect size, and heterogeneity/publication-bias plan.
GCB reviewers probe whether the design can bear the weight of the global-change claim. Use this to locate the soft spot before a referee does and to choose the strengthening move.
| Design soft spot | Reviewer phrasing | Strengthening move |
|---|---|---|
| Treatment confounded with unit | "Pseudoreplication" | Replicate at the inference level; state the unit |
| Dose far above realistic change | "Unrealistic forcing" | Add a realism gap statement or a dose gradient |
| Space-for-time as causal | "Gradient is not an experiment" | Add covariates or a confounder model |
| Single model run | "No structural uncertainty" | Move to an ensemble; partition uncertainty |
| Unstated search protocol | "Synthesis not reproducible" | Pre-register a PRISMA-style protocol |
A team plans an open-top-chamber warming experiment to test a soil-respiration feedback. A weak design warms one large chamber and samples it 30 times, then treats those as 30 replicates — pseudoreplication a GCB referee will flag immediately. The strengthened design uses six warmed and six control plots (illustrative n), warming each by an ecologically realistic +2 C rather than +6 C, and pre-commits to a mixed model with plot as the random unit. Power analysis (illustrative) suggests this detects a 15% efflux change. The realism gap and the scaling limit to ecosystem level are stated up front. Numbers illustrative.
process-model test of the mechanism.
uncertainty, and bound rather than assert the larger claim.
driver.
【Design family】experiment / gradient-observational / model / synthesis
【Driver & response】manipulated/measured at what scale
【Replication & unit】level of inference; pseudoreplication ruled out? [Y/N]
【Realism / confounding】dose-duration realism or confounder plan
【Uncertainty plan】measurement + model + scenario
【Next】gcb-data-analysis../../resources/external_tools.md — experimental, observational, and modelling toolchains../../resources/official-source-map.md — GCB scope (molecular-to-biome, aquatic/terrestrial)~30 seconds. Free. No account. Every finding cites a rule and a line of evidence.