Contract validation
You bring a product and a claim. We design, execute, analyse and report. You own the data. Most common route for companies preparing a launch, a dossier or a funding round.
Research
Research programme and methodology
The design rules, measurement endpoints and analytical standards that every trial at the center is held to — published openly, because a method you cannot inspect is not a method.
Approach
Agronomic questions, answered with clinical discipline
Field agriculture is statistically hostile. Soil varies across a few metres, weather is uncontrolled, pest pressure is uneven, and biological responses are noisy. A single treated plot next to a single untreated plot tells you almost nothing, however striking the photograph.
Our answer is to borrow the structure of clinical trial methodology and apply it to crops: replication sufficient to detect the effect size that matters, randomised allocation to remove positional bias, concurrent controls, pre-declared endpoints, and analysis planned before the data exists.
What we measure
Endpoints are selected for the claim being substantiated, not for convenience. A germination claim and a yield claim require different designs, different durations and different plot counts — and we will say so at the proposal stage rather than after the season.
| Endpoint | Stage | Method | Typical replication |
|---|---|---|---|
| Emergence rate & timing | Sowing to establishment | Daily seedling count per plot against sown total | 20–40 seeds / arm |
| Early vigour | Establishment | Shoot length, leaf count, visual vigour index | 15–30 plants / arm |
| Root architecture | Vegetative | Destructive sampling, washed root imaging, dry mass | 6–12 plants / arm |
| Biomass accumulation | Vegetative to flowering | Fresh and oven-dry mass, shoot and root separated | 10–20 plants / arm |
| Stress tolerance | Imposed stress window | Relative water content, chlorophyll index, recovery scoring | 15–30 plants / arm |
| Fruit set & count | Reproductive | Per-plant fruit enumeration across the harvest window | Per-fruit datapoints |
| Yield & quality | Harvest | Marketable mass, grading, calibre distribution | 4–8 plots / arm |
| Soil response | Multi-season | Organic matter, structure, respiration, nutrient status | Composite / block |
Design standards
The rules we do not bend
Replication and statistical power
Plot count is derived from the effect size the claim requires, the variance observed in our historical baseline for that crop, and a target power of 0.8 at α = 0.05. If the budget will not support the replication the question needs, we say so before contracting rather than delivering an underpowered study that cannot distinguish signal from noise.
Where a pilot is genuinely exploratory, we label it as hypothesis-generating and design the confirmatory stage separately. A pilot is not a claim.
Randomisation and blocking
Treatments are allocated by randomised complete block design, with blocks oriented against the dominant field gradient — typically drainage or shading. Plot positions are generated before establishment and recorded in the protocol annexe.
Guard rows separate treatment plots where drift, runoff or root interference is plausible. For irrigated trials, arms are hydraulically isolated.
Control management
Control plots receive an identical volume of carrier applied on the same day by the same operator. Where a product is applied in water, controls receive matched water — a control that is simply left alone confounds the treatment effect with an irrigation effect.
Both arms receive the same background agronomy. Where the trial specifies no synthetic fertiliser or pesticide, that constraint applies equally to both arms and is recorded in the log.
Application records
Every application is logged against date, time, operator, dose, dilution, volume per plot, equipment and prevailing weather. The log is delivered with the report. Deviations from protocol are recorded as deviations, not silently absorbed.
Analysis and reporting
Analysis follows the plan declared in the protocol. We report effect sizes with confidence intervals alongside significance tests, because a p-value alone tells a commercial reader nothing about whether an effect is worth paying for.
Where assumptions of normality or homogeneity of variance fail, we report the diagnostic and the alternative test used. Multiple endpoints are handled with declared correction. Post-hoc exploration is reported as post-hoc.
Handling of null and adverse results
A trial that shows no effect is a completed trial, not a failed one, and it is reported with the same detail as a positive result. Phytotoxicity, yield depression or quality loss are reported immediately on observation, not held to the final document.
Collaboration models
Four ways organisations work with us
Joint research programme
Shared design and shared intellectual contribution, typically with a university or institute partner, leading to co-authored publication. Costs and credit are agreed in advance.
Multi-site network trials
The same protocol run here and at partner sites in other climates, producing multi-environment data from a single contract and a single analysis.
Due diligence assessment
Commissioned by investors or acquirers to independently examine whether an existing evidence base supports the claims being made for a technology.
Get started
Bring us a question the field can answer
Send the claim you need to substantiate and the season you are working toward. We will tell you what design it requires — and whether it is achievable in the time you have.