Experiments

When sources conflict, are silent, or a production step is central at small scale but undocumented, the project runs original experiments. Experiment results become evidence like any other source, and are cited from crop pages by their record id (e.g., EX-0001).

This page is the design guidance; the operational templates live in templates/experiment.md and experiments/.

When to run an experiment

Prefer experiments when:

Sources genuinely conflict under comparable conditions, or
No source addresses the question at small scale / by hand, or
A claim is central and would change advice if wrong, or
A “Needs experiment” open question is worth resolving.

Experiment time is limited; pick questions that matter for the human-scale manual case.

Minimum design standards

Even an informal backyard trial must record enough to be interpreted:

One question at a time — vary one factor, hold the rest constant.
Controls and/or comparison — a treatment needs something to be compared against (untreated, or the source’s suggested method).
Replication where feasible — few, but at least some. A single plant is a demonstration, not a result. Note explicitly when replication is impossible.
Simple metrics that matter — e.g., grams of usable seed per m², time to thresh 1 kg, seed loss percentage, moisture reached. Prefer outcomes along the real chain (usable product), not just “plant grew.”
Weather/location log — date, location, temperatures (min/max), rainfall, photoperiod.
Fixed units and conversions — follow units.md.

The record

Use templates/experiment.md. A completed record must include:

Question and hypothesis
Method (as-run, not as-planned — note deviations)
Materials (cultivar, seed source, equipment)
Conditions log
Measurements and raw data
Analysis (means, spread; be honest about small n)
Resulting evidence level with rationale, and proposed claims
Known limitations (“n = 3, one season, dry year — not yet Strong”)

File it under experiments/records/EX-####-<short-topic>.md. If the experiment spawns a claim strong enough for structured data, update data/crops/ and cite EX-####.

Ethics and honesty

Do not cherry-pick runs. Report all replicates, including failures.
Do not overstate: a single backyard season is Limited. Several seasons/locations → Moderate+.
A failed experiment is a result (e.g., “hand threshing of dry chia took ≥2 h/kg and lost 8% — this needs a better method”) — record it; it may be more useful than a success.

Safety and practical limits

Follow local rules on cultivation.
Do not encourage risky storage of flammable oilseed residues or unsafe tools. Note any hazards in the record.
Experimental plots may be small, but label them as X m² so results remain comparable via units.md.