Help us improve
Share bugs, ideas, or general feedback.
From grimoire
Structures biological experiments with controls, randomization, blinding, and power analysis to produce valid reproducible results. Uses GLP and Fisher principles.
npx claudepluginhub jeffreytse/grimoire --plugin grimoireHow this skill is triggered — by the user, by Claude, or both
Slash command
/grimoire:design-biological-experimentThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Structure a biological experiment with controls, replication, and statistical power to produce valid, reproducible results.
Designs detailed experimental protocols for validating research hypotheses, including variables, controls, power analysis, timeline, and expected outcomes.
Provides Python code patterns for reproducible experiments: random seeds, environment logging, train/test splits, cross-validation, A/B testing, and power analysis. For ML/statistical designs.
Plans physics experiments that isolate causal variables using controlled design, randomization, blocking, and replication.
Share bugs, ideas, or general feedback.
Structure a biological experiment with controls, replication, and statistical power to produce valid, reproducible results.
Adopted by: OECD member nations (GLP compliance), NIH-funded research programs, Cold Spring Harbor Laboratory, peer-reviewed journals requiring ARRIVE/CONSORT reporting.
Impact: Fisher's randomized block design reduced experimental error by 30–50% in agricultural trials; GLP-compliant studies have a 60% lower rate of retraction vs. non-compliant studies (Fanelli 2012).
Why best: Randomization eliminates selection bias; replication separates signal from noise; blinding prevents observer bias — together these are the minimal conditions for causal inference in biology.
Sources: OECD GLP Principles (ENV/MC/CHEM(98)17); Fisher (1935); Cold Spring Harbor Protocols experimental design series.
State the hypothesis — write a single falsifiable statement in IF-THEN-BECAUSE form (e.g., "If gene X is knocked out, then cell proliferation will decrease by >20%, because X activates the MAPK pathway").
Identify variables — list independent variable (what you manipulate), dependent variable (what you measure), and all confounders you will control.
Define controls — include a positive control (known outcome), negative control (no treatment), and vehicle control (solvent only) for every experimental group.
Calculate sample size — use power analysis (α=0.05, β=0.20, effect size from pilot or literature) before starting; target ≥80% power. See calculate-statistical-power.
Assign randomization — randomly assign subjects/samples to groups using a random number table or software (R, Python) to prevent systematic bias.
Plan blinding — blind the experimenter to group assignment during measurement wherever feasible; use coded labels.
Write the protocol — document each step with exact reagent concentrations, instrument settings, timing, and acceptance criteria for data quality.
Specify statistical analysis — pre-register the primary statistical test, multiple-comparison correction method, and exclusion criteria before data collection.
Execute and record — record all deviations from protocol in a lab notebook contemporaneously; photograph key results.
Validate reproducibility — replicate the key experiment ≥3 times on separate days (biological replicates, not just technical replicates).