Proof before prediction: why every farmland figure must carry its source

Farmland decisions are still often made on a declared production history and a site visit. The public data that could test those claims exists, but it is scattered across land registries, statistics, soil, water, climate and regulation, and its provenance and legality are rarely explicit. A number without its origin cannot be challenged, so it cannot be trusted either.

• Count: how many observations stand behind the value.
• Coverage: which area, crop and period it describes.
• Method: measured, modelled or inferred, and by which documented method.
• Date: when the inputs were read.
• Confidence: an explicit band, not a single headline number.

When a value cannot be produced, Phoeniks says so and says why: a source that does not cover the area, a provider that is not configured, a model that is not provisioned. The empty state is designed, not hidden. A zero or an average put in place of a missing value would look complete and mislead the committee that relies on it.

Every figure is labelled measured, modelled, inferred or untested. A reviewer sees at once what was observed and what was computed, and where the file needs another check before a decision.

The question moves from “do we believe this number?” to “what does it rest on?”. A model-risk function, an auditor or a co-investor can follow the figure back to its sources and method. Phoeniks does not publish headline accuracy claims detached from crop, geography and season, because such a claim cannot be tested.

• Phoeniks methodology page: https://phoeniks.org/en/trust/methodology
• Phoeniks release notes, V1 (17 October 2026): https://phoeniks.org/en/release-notes/v1