Research protocol (summary)
The canonical text is research-protocol.md in the lab's operating repository (private). This page is a summary; the principles are identical to the canonical text.
Principles
- Structure claims and evidence before writing articles.
- Store short summaries that lead back to the original text, not collections of quotations.
- Do not treat human sensory tests, receptor assays, animal or cell studies, and instrumental analysis as equivalent.
- Store not only support, but also counter-evidence, limiting conditions, and open questions.
- Do not identify a plant species from a photograph alone.
- Do not assume that one dish name or plant name refers to one thing.
- Do not read a use in food culture as proof of safety or of a health effect.
- When calling a search "comprehensive", publish the databases, search strings, period, and inclusion flow.
Confidence labels on claims
- established — broadly consistent across multiple independent, high-quality studies
- supported — direct evidence exists, but the subjects or conditions are limited
- preliminary — few, small, non-human, or indirectly measured
- contested — leading studies do not agree
- contextual — history, culture, classification, regulation: sources of a different kind from experimental causation
- observation — records such as the chronology of a literature or field observations, in which causes are not separated; not turned into general rules
- unverified — general-knowledge-level culture, classification, or dish composition that has not yet been checked. The card states that no source is connected, and the item is treated as a queue for later verification. Numbers, mechanisms, safety, and health effects are not recorded at this level, and recipe candidates carry no quantities, temperatures, or times
Hypothesis box
Unverified questions, things people say, and oral or transmitted lore are stored separately from claim cards that state facts. Editorial hypotheses are written as questions and never dressed up as someone's statement. For "it is said that" stories, the origin is recorded; when unknown, the entry shows "origin not recorded". Confirming in a source that a story exists is not the same as confirming that its content is true.
Check status of sources
The ledger distinguishes "abstract checked" from "full text checked". Numbers, sample sizes, and statistics are never generalised from a source checked only by its abstract.