Greenhouse data tells you whether an effect exists under controlled conditions. Field data tells you whether that effect survives contact with soil heterogeneity, weather variation, existing microbial communities, agronomic practices, and all the other variables that a greenhouse trial holds constant by design. For biostimulants targeting stressed soils, the gap between greenhouse signal and field result is one of the most documented failure points in the agricultural biology literature. Understanding why that gap exists and how to design trials that give you useful data even with it present is the practical problem we are working through now that NB-001 is in greenhouse validation.
This piece is about trial design principles, not our results. We do not have field data yet. We have greenhouse data from NB-001 and we are in the process of designing the field trial framework that will follow. Writing about that design process forces us to think through the decisions explicitly, which is useful independent of whether the article is useful to anyone else.
The Variance Problem in Stressed Soils
The fundamental statistical challenge in field trials on degraded saline land is variance. Saline soils are not uniformly saline at the meter scale. Salt accumulates in patterns driven by topography (slight depressions accumulate more salt than elevated positions), irrigation history (parts of a field that received more irrigation carry more accumulated salt), drainage patterns, and soil texture heterogeneity. The electrical conductivity across a single 10-hectare field on degraded irrigated land in Mendoza province might range from 3 dS/m to 12 dS/m depending on where within the field you measure.
This means that if you assign treatment and control plots without accounting for the salinity distribution within the field, you may end up comparing your biostimulant treatment in higher-salinity zones against your control in lower-salinity zones, or vice versa. The variance from soil heterogeneity will swamp any treatment effect, particularly if the treatment effect is modest in absolute terms, which it is likely to be at early stages of development.
Blocking on soil electrical conductivity is the standard approach: you measure conductivity across the field in a systematic grid before assigning plots, divide the field into zones of similar conductivity, and then randomize treatment assignments within each zone. This is electroconductivity-stratified randomized block design, and for field trials on heterogeneous saline land it is close to mandatory if you want interpretable results.
Plot Size and Replication Trade-offs
For a small research group working without institutional field station infrastructure, there is a hard resource constraint on plot size and replication. More plots mean more inoculant, more sampling labor, more laboratory analysis. Fewer, larger plots reduce statistical power but reduce logistical complexity. The right trade-off depends on what effect size you are trying to detect.
Expected effect sizes for biostimulant treatments on stressed crops in the peer-reviewed literature are generally modest: yield improvements in the 5-20% range under moderate stress are considered biologically meaningful results, not exceptional ones. To detect a 10% yield difference with 80% statistical power at p less than 0.05, with realistic within-plot variance for a stressed cereal crop, you need somewhere in the range of 4-6 replicate blocks. With smaller plot sizes (5-10m x 10m), this is logistically achievable; with larger plots the cost per replicate increases substantially.
We are currently thinking about minimum usable plot sizes for our application method. Seed inoculation can be applied uniformly at planting. Soil drenching at planting requires more inoculant volume but provides more reliable delivery for certain crops where the seed coating does not provide sufficient inoculum load. The application method determines the minimum plot size for a valid treatment, because below a certain plot area the edge effects from untreated adjacent plots contaminate the treatment outcome.
What to Measure and When
Yield at harvest is the headline metric, but it is insufficient as the only measurement. We detailed the intermediate measurement set in an earlier piece on biostimulant versus fertilizer comparisons: stand establishment, root architecture at early vegetative stage, leaf area index and chlorophyll content through the season, and harvest index alongside grain yield. The key addition for field trials on stressed land is soil and environmental monitoring throughout the season.
Soil conductivity measurements at each plot at planting, mid-season, and post-harvest create a record of the actual stress conditions experienced by each treatment and control plot. Irrigation events and rainfall during the season need to be tracked. Temperature and frost event records for the site need to be maintained. Without this environmental record, you cannot interpret a negative result: a control plot that happened to sit in a lower-conductivity zone and a treatment plot in a higher-conductivity zone would produce data showing the biostimulant had no effect or a negative effect, when the reality is that the stress conditions were not comparable.
Microbial community sampling at planting, mid-season, and harvest from treated versus control plots provides a separate line of evidence about whether the inoculant is establishing and persisting in the rhizosphere. We run quantitative PCR for the inoculant strain in root zone soil at each sampling point. If recovery drops to near-zero by mid-season, the treatment effect in the crop data reflects only the early-season colonization; if recovery is maintained through the season, the effect is more likely to be ongoing. This information affects how we interpret the crop response and how we would modify the inoculant formulation or application rate for subsequent trials.
Multi-Year and Multi-Site Considerations
A single-year, single-site trial is necessary but not sufficient to support a decision about advancing a candidate to development. Year-to-year variation in weather, and particularly in the occurrence of stress events during critical crop development windows, means that a positive result in one season at one site could reflect an unusually favorable match between the trial conditions and the candidate's mechanism of action. Conversely, a negative result in one season might reflect an unusually low-stress growing year where the biostimulant's effect was not needed.
Multi-site and multi-year replication is the standard requirement for an agricultural claim that is worth making. For a small early-stage group, committing to a three-year, three-site trial is a significant resource decision. We are thinking about how to structure a phased approach: one site for two consecutive seasons as a primary validation, with expansion to additional sites conditional on observing a consistent positive signal at the primary site. This is not the ideal statistical design, but it is a realistic sequencing for a group our size working without major institutional backing.
The Honest Uncertainty We Carry Into This
We know the greenhouse result is promising. We also know that the history of microbial agricultural products includes many examples of strong greenhouse results that did not replicate in field conditions. The most common failure modes are poor persistence of the inoculant in competition with native rhizosphere communities, stress conditions at the field site that differ structurally from the conditions tested in the greenhouse, and effect sizes too small to detect given the variance of field trials on heterogeneous land.
We are not claiming these failure modes will not apply to NB-001. We are designing the trial to give us the best possible chance of detecting a real effect if it is present, and to give us interpretable data regardless of the outcome. If NB-001 fails in well-designed field conditions, that is important information that shapes everything we do next. If it succeeds, that is the foundation we need to have a credible conversation with agricultural input partners about development. Either result is useful, which is the correct orientation for a trial at this stage.