Every reading knows its plot, its person and its fix
Fieldbase treats a trial as ground: plots are polygons with real dimensions, a phone in the field knows which one it is standing in, and every observation carries who took it and where the value came from. The analysis is built to be defended, not just displayed.
Designs onto real geometry
Author a completely randomised, randomised complete block, Latin square, split plot or factorial design and place it with a seeded randomisation you can reproduce, or import the fieldbook your statistician already made and bind it onto the grid.
Provenance on every reading
Each observation records the operator and the source (GPS, manual, instrument), and plots keep a stable identity, so inserting a range never slides a season of data onto the wrong plots.
Analysis that refuses to guess
Analysis of variance with mean separation, scoped to the design you declared. When the design cannot support an honest error term, Fieldbase says so and stops, rather than printing letters with a disclaimer.
Instruments straight into the record
Scales, grain gauges, moisture meters and NIR analysers over serial, with the parse defined as data you can adjust. Barcode label sheets tie samples back to the plot's stable id.
Assessments walked in order
Guided rounds step through the plots in walk order, trait by trait, with range checks that warn and never block. The person in the plot is the one who can tell a miscount from a discovery.
Results in your pipeline's format
One row per plot in XLSX or CSV, a BrAPI v2 hand-off for breeding pipelines, and shapefile, ISOXML, KML and GeoJSON for anything that thinks in geometry. Export is never gated.
What changes through a trial year
Pegging out plots from a printed plan and hoping the numbers match
The grid is generated from the dimensions, numbered to match the fieldbook, and placed against the boundary with typed clearances.
Transcribing paper fieldbooks and instrument tickets in the evening
Observations land against the plot as they are taken, with operator and source attached; instrument readings arrive over serial already parsed.
A statistics package that trusts whatever reached the spreadsheet
The analysis knows the declared design, tests before it separates means, and refuses when the structure cannot support the comparison.
One map holds the whole farm and every trial on it: boundaries, grids and plots are layers of the same view, and a trial can span fields without changing its identity. The research programme (experiments, traits, assessment rounds, dashboards) lives alongside in the same file, so the plot a machine drove in June and the value analysed in September are the same record, not an export of one another.