Methodology
How the analysis works, and what it cannot tell you.
Folia produces a read of a property from public data. It is useful because it is specific, and because it is honest about its own limits. This page describes both. Start with what it does not do.
What Folia does not do
It does not tell you what to plant. The analysis reads the ground, ranks what will matter most, marks what is still uncertain, and names what to confirm on site. What to plant remains yours. That depends on your market, your capital, and your appetite for management, none of which a dataset can see.
It does not replace walking the ground. Modeled values are modeled. The report ships field worksheets precisely so you can check them against your own measurements, and it names the places where a soil pit or a county office is the only real answer.
It does not guess. Where a source cannot answer for your property, the report prints that the value is unknown. It does not fill the gap with a plausible number.
No language model in the reasoning path
The engine is deterministic and rules-based. Every value comes from a named public dataset or a published method, and the prose that explains it is templated from those values. Run against the same inputs, the assessment reproduces exactly.
There is no language model anywhere in the path. Nothing here asks a model what grows on your land. That matters more than it used to: a tool built on an AI model produces fluent, confident, unverifiable output, and the difference is not a matter of quality. It is whether the number came from anywhere. Folia's come from the datasets named below.
Every value carries a grade
Folia never presents an estimate as a measurement, and treats an unknown as a real answer rather than a blank. Each finding is marked with how firmly it is known.
| Grade | Meaning |
|---|---|
| Sourced | Returned directly by an authoritative public dataset for your property. Reads as firm. |
| Derived | Computed arithmetically from measured data; carries the precision of its source grid. |
| Modeled | Produced by a statistical surface or interpolation. Correct at its cell scale, approximate at property scale. |
| Estimate | A range judged from regional reference data. Always presented as a band, never a point. |
| Unknown | The underlying source could not answer. The gap is stated, never filled. |
On the pages that weigh a variety against the land, a second reading appears: an Assessment Confidence of High, Moderate, or Low. That grades how strongly Folia stands behind the interpretation, separately from the data source beneath it. Neither one tells you what to plant: the interpretation is Folia's, the decision stays yours. The grades are visible in the report itself, on the values they apply to, not an internal note.
Where the data comes from
The assessment is one query fanned out to the public services below, run in parallel. Each degrades independently: a failed source is reported as unknown for that layer and never substituted.
| Layer | Source |
|---|---|
| Elevation, slope, aspect, and 1 m LiDAR canopy | USGS 3DEP |
| Soil units, drainage, and water-holding capacity | USDA NRCS SSURGO |
| Climate normals, growing degree days, and frost window | PRISM, served via ACIS |
| Winter hardiness | USDA Plant Hardiness Zone Map (2023) |
| Flood exposure | FEMA National Flood Hazard Layer |
| Wetlands | State wetland inventories where available, otherwise the National Wetlands Inventory |
| Wildfire hazard | USFS Wildfire Hazard Potential |
| Variety traits and thresholds | Folia Crop Intelligence Library |
The crop library and its limits
Variety thresholds come from the Folia Crop Intelligence Library, which is built and maintained rather than scraped. Six crop classes are live: grapes, apples and pears, stone fruit, berries, citrus, and perennial flowers.
Coverage is uneven, and the report says so rather than hiding it. A short list of suitable varieties can mean the ground is difficult, or it can mean the library is thin for your region. Those are different answers, and the report distinguishes them. For warm-zone crops in particular, sparse modeling can produce false negatives, so a variety that “does not fit” there is weaker evidence than the same result for a well-modeled crop. Where regional coverage is limited, it is labeled as limited rather than presented as a complete result.
Grades move when the evidence moves
A confidence grade is only worth something if it can be revised downward. When new published evidence contradicts what the library holds, the affected values are regraded and the change is recorded.