Research methodology

Research Methodology

This page describes the research loop used to turn scattered sources into structured project knowledge.

The research loop

1. Frame the question. Per crop, per topic. Example: “How is chia seed separated from the plant at small scale, and at what loss?” (See docs/crops/chia/harvesting.md for how questions attach to pages.)
2. Survey source types in rough priority order:
   • Academic literature (peer-reviewed, but read critically — see source-quality.md)
   • University and agricultural-extension publications
   • Government resources
   • Books (serious, cited works)
   • Commercial farming information (flag the commercial interest)
   • Practical grower demonstrations (videos, forums, blogs)
   • Original experiments (ours, experiments/)
3. Extract claims with context. Every claim leaves the source together with: conditions (cultivar, location, date, density, irrigation, fertilization, design), the evidence level, and the source id. Never strip a yield number from its conditions (principle: preserve context).
4. Register the source. File a source note in research/<type>/ using the templates, assign a stable id, assess quality and relevance (see source-quality.md).
5. Write or update the narrative page. Claims land in the appropriate crop/topic page, tagged with evidence level and source. Conflicting claims both stay, with the disagreement noted (evidence.md).
6. Update the structured data. When a claim is strong enough to represent a “typical” or “range” value, mirror it into data/crops/<crop>.yaml with provenance. Unresolved values stay null/unknown.
7. Record open questions. Anything the sources cannot answer goes to open-questions.md, optionally as a candidate experiment.

Scope decisions

• One crop at a time for depth, others for comparison. Chia is the first complete case study. Amaranth, sesame, sunflower, and dry beans follow; comparisons matter (PRD §8).
• Initial scope: home and small-scale agriculture. Commercial practice is documented where it illuminates the small-scale question (e.g., what a commercial chia grower does that a home grower cannot), but small-scale feasibility is the lens.
• Micro-decisions get recorded. When the research reveals that a “fact” depends on a micro-condition (e.g., photoperiod sensitivity differing by cultivar), that nuance belongs in the narrative and, if possible, the structured data.

Handling source types honestly

Source type	Strength	Weakness	How we use it
Academic paper	Rigor, controls, quantitative	Researcher/commercial framing; narrow conditions	Primary quantitative claims; always with context
Extension bulletins	Practical, regional, actionable	Region-specific	Practical recommendations; note region
Government	Authoritative for their jurisdiction	Jurisdictional	Regulatory/policy context; statistics
Books	Synthesized, historical depth	May be dated	Background and synthesis
Commercial	Real-world production experience	Marketing bias	Practice reference; flag bias
Growers/YouTube	Human-scale practical detail	Uncontrolled, variable quality	Techniques, manual methods; lowest evidence default

The same claim from different source types may earn different evidence levels. That is by design.

Practical evidence vs. scientific evidence

These are not interchangeable (principle 6.6):

• A paper may show optimal N application.
• A grower video may show how to thresh 8 kg of chia in a backyard.

The narrative labels each claim’s type — e.g. scientific vs practical — in addition to its evidence level, so a reader can weigh them separately. See evidence.md.

Original experiments

When sources conflict, are silent, or a claim is central at small scale, an original experiment may be warranted. Experiments follow experiments/ conventions and are recorded with the template in templates/experiment.md. Design guidance in experiments.md. Outcome is reported as Needs experiment → experiment + result.

When is a topic done?

A topic is “done” when: every sub-question in the page has an answer or an explicit Unknown, disagreements are recorded with context, structured data is in sync, and open questions are filed. “Done” is a milestone, not a stop — new sources keep landing as updates, not rewrites.