Field validation

Review → Field validation compares automated plot signals with independent reference evidence on a dedicated page. Shipment reviews continue to assess individual shipments and their evidence.

1. Create a review run. Choose a name and a target of 3–300 plots. Random selection is stratified by signal group and uses previously screened plots. If fewer eligible plots exist, the run contains fewer plots. The screening link opens plot screening runs.
2. Review independently. Select a run and then a plot on the left. Its assessment and reference evidence appear on the right. Before your first assessment, the automated result and other reviewers' labels remain hidden. Choose conversion, no conversion or uncertain; provide an observation date, at least one independent reference document and a rationale of at least 30 characters. Saving reveals available results. Further assessments create new recorded entries rather than overwriting earlier reviews. Editing requires compliance.write; viewing requires compliance.read.
3. Compare results. Metrics distinguish confirmed signals, unconfirmed signals, missed signals and confirmed no-signal results. Two distinct reviewers must agree on a definite reference verdict before a plot can be evaluated. Uncertain or conflicting references and non-definite automated signals remain unresolved. The JSON report contains the selected run and sample data available to you.

These are pilot metrics for the selected sample, not a population accuracy guarantee or shipment approval. A stratified sample requires appropriate weighting before generalisation.