Evidence

Evidence

Every substantive claim in this project carries an evidence level. Unknown is a valid and respectable answer; fabricated certainty is the worst outcome (PRD §6.2).

The evidence scale

Level	Meaning	Example phrasing
Known	Established fact with broad consensus across multiple independent sources (e.g., basic biology, chemistry, well-established agronomy)	“Chia is a short-day plant.”
Strong	Multiple consistent, high-quality sources (e.g., replicated studies or several independent practical reports) with matching conditions	several studies report…
Moderate	A few good sources, or one high-quality source; some consistency	a field trial reported…
Limited	One or few low-to-medium quality sources; limited confirmation	a single grower report…
Anecdotal	Unverified personal or community claims; plausible but unchecked	a forum user reports…
Conflicting	Genuine disagreement between sources under comparable conditions — record both, don’t average (see below)	Study A reports X; Study B reports Y
Unknown	No reliable information found	not yet known
Needs experiment	Question is answerable but the evidence is missing or conflicted; an original experiment is warranted	a small-scale trial is needed

Two extra dimensions on every claim

1. Evidence type: scientific (controlled study, formal measurement) vs practical (grower demonstration, manual technique) — these are not interchangeable (PRD §6.6).
2. Context: conditions under which the claim holds — cultivar, location, climate, date, density, irrigation, fertilization, design, source. A number without context is misleading (PRD §6.4).

How to tag claims

Preferred inline style, kept compact:

Reported yield: 1,200 kg/ha — Moderate · scientific — cultivar X, dryland, Spain, 2019 (S-0117).

Agreed abbreviations: K Known · S Strong · M Moderate · L Limited · A Anecdotal · C Conflicting · U Unknown · E Needs experiment. Use the abbreviation in structured data (evidence: M), the long form in narrative prose.

Structured data mapping

data/crops/*.yaml values carry the evidence level of their source claim. A field is either:

• a value + evidence + provenance (e.g., optimal_min_c: { value: 16, evidence: M, source: [S-0117] }), or
• null / unknown when there is none.

Tools (comparison UI etc.) must never display an evidence-less value as fact; they should render the level and suppress or flag unknown fields.

Preserving disagreement

Different results under different conditions are not contradictions — they are information. Differences can arise from cultivar, temperature, photoperiod, soil, fertilization, irrigation, density, planting date, rainfall, disease, harvest timing, or experimental design (PRD §6.3).

Rules:

1. Record each result with its conditions.
2. Note the likely source(s) of difference explicitly.
3. Do not average conflicting numbers into a single “typical” value until you understand the conditions that caused the difference.
4. When you do publish a typical/range value (e.g., in structured data), state the conditions it covers and list the disagreement.

Evidence level of a page vs. a claim

Pages are not assigned a single evidence level — they are containers of claims, each with its own level. An overview page can summarize (“optimal germination temperature ≈ 18–25 °C”), but must link to the tagged claims underneath.

Escalating from one level to another

A claim’s level changes only through new evidence:

• Anecdotal → Limited when independently corroborated (a second grower, same technique).
• Limited → Moderate when confirmed by a decent study or systematic practical reports.
• Moderate → Strong when replicated under different conditions with consistent results.
• Conflicting → Strong only when the cause of disagreement is identified and the conditions are mapped.
• Unknown → Limited via an original, recorded experiment (experiments/).

Every escalation should be visible in git history with the added source.