Research and Editorial Methodology
Trust should be inspectable. This page describes exactly how a claim gets into the catalog, what it is allowed to say, and how the planner uses it.
Last updated August 24, 2026
The evidence rule
Popular companion charts mix several different concepts (intercropping, trap crops, insectary flowers, rotation, allelopathy, plain competition) and present all of them as universal fact. The catalog keeps them apart. "No verified relationship" is not the same as "neutral" or "safe"; it is an explicit finding that we looked and found no usable evidence. A negative label requires a credible harmful interaction or an obvious agronomic conflict, not repetition across unsourced charts.
Six statuses
| Status | Meaning | Planner behaviour | Records |
|---|---|---|---|
| Supported benefit | Direct or strong synthesized evidence applies to this crop pairing and context. | Positive score when the plan matches the documented conditions. | 3 |
| Conditional benefit | Credible evidence, but the benefit depends on timing, arrangement, pest, cultivar, climate or management. | Positive only when the stated conditions match; otherwise informational. | 25 |
| Traditional claim | Repeated gardening tradition or credible field observation without adequate direct evidence. | Never a hard rule; may appear as an optional experiment. | 43 |
| Avoid or separate | Credible negative evidence or a clear physical or cultural incompatibility. | Hard constraint only for serious, applicable conflicts; otherwise a weighted warning. | 7 |
| Conflicting evidence | Good sources disagree, or outcomes differ materially by context. | Both findings are shown; the planner neither rewards nor blocks automatically. | 1 |
| No verified relationship | The editorial review found no usable evidence of benefit or harm for this pair. | Zero score and an honest unknown state. Spacing, light, water and timing are evaluated normally. | 40 |
Five evidence tiers
| Tier | Requirement | Planner weight | Records |
|---|---|---|---|
| Tier A | Systematic review or meta-analysis, or multiple relevant replicated field studies with consistent results. | 1 | 1 |
| Tier B | One or more relevant controlled field studies or an authoritative synthesis tied to the crop, effect and arrangement. | 0.7 | 16 |
| Tier C | Indirect evidence: greenhouse or pot work, a related crop or species, or credible extension guidance with important uncertainty. | 0.35 | 19 |
| Tier D | Traditional knowledge or structured community observations without a controlled comparison. | 0.1 | 43 |
| Unknown | No source meeting the inclusion rules. | 0 | 40 |
Constraints enforced at build time: supported requires tier A or B; avoid requires tier A–C and a hard constraint requires A or B; traditional is always tier D; unknown is always unknown; allelopathy claims must state whether their evidence is laboratory, pot or field; and opposing records for the same pair must reference each other.
Direction and context
Every record names plant A, plant B and a direction (A→B, B→A or mutual). "A affects B" never implies "B affects A". Every record also states the production context (home garden, field, protected), the spatial arrangement (mixed, row, strip, perimeter, undersown, relay), distance or density, relative planting time, the overlap requirement, management requirements, applicability and limitations. A relationship can influence the planner only when the plan matches that context.
Research procedure
- Normalise accepted scientific names (USDA PLANTS).
- Define the exact claimed outcome: yield, pest abundance, pest injury, disease, pollination, space efficiency.
- Search systematic reviews and crop-specific trials first; then extension syntheses and manuals.
- Record crop, cultivar, climate, arrangement, density, timing, comparison and outcome.
- Check for contradictory studies and for costs (yield, labour).
- Judge whether field findings transfer to a home garden.
- Write a limited claim that matches the evidence.
- Second-reviewer approval for supported and avoid records.
- Version the record, log the change, schedule review in 18–24 months.
Why community reports do not change tiers
A gardener's report is an uncontrolled observation from one site and season. It is valuable as a prompt for research and as a warning, and it is shown in a separate, clearly labelled area with sample size and context. It never changes an evidence tier or a planner score. See the community guidelines.
Planner weighting and its limits
For each applicable relationship the planner computes edge_score = direction × evidence_weight × context_match × time_overlap × distance_match × management_match × user_priority. Resource competition, shade, incompatible water or fertility, shared disease risk and access problems are separate penalties. Every component stays visible in the reason ledger; nothing is collapsed into an unexplained number. The weights are configurable editorial settings, tested for sensible behaviour, not biological truth. A weak companion claim never overrides spacing, season, light, water or rotation constraints.
Constraint order: safety and legality → physical feasibility → season feasibility → resource compatibility → rotation → companion relationships → user preferences.
Review frequency and change log
Every record carries a reviewer, a review date and a version. Records are scheduled for re-review every 18–24 months or when a correction is accepted. The change history is the repository's commit log; substantive content changes update the "last reviewed" date on the page.
Authors, reviewers and conflicts of interest
At launch every record is marked "Garden Planner editorial team (pending independent horticultural review)". No external reviewer is named until a review has actually taken place. Nobody involved receives payment from seed, tool or input companies. See authors and reviewers.
Current coverage
63 plants, 119 relationship records, 33 sources. Coverage is deliberately uneven: cabbage, broccoli, lettuce, tomato and the squash family have real trials behind them; many herbs and flowers have only tradition, and say so.
Known limitations
- Most field evidence comes from farm-scale trials; transfer to a small bed is a judgement recorded in each record's applicability notes.
- Regional calendars for limited-support regions are estimates derived from frost dates and need local confirmation.
- Yield figures are wide ranges for planning, not predictions.
- Two meta-analyses are cited by DOI as given in the project brief with a note to verify the full titles; see the sources library.
Data and provenance are downloadable from the companion dataset page. The editorial protocol is published in the repository as docs/editorial-protocol.md.
