GardenSeerLibrary
Methodology

Useful garden intelligence starts with an honest record.

GardenSeer is designed to preserve the difference between what was observed, what may be inferred, what is forecast, what is recommended, and what was later confirmed.

Evidence levels

Every statement should be understandable in terms of the record beneath it. GardenSeer uses the following distinctions so growers can judge an answer instead of receiving a black-box conclusion.

ObservationSomething directly seen or recorded, tied to a source and time.
Grower actionAn action actually taken, recorded without assuming its effect.
CorrelationChanges that occurred together, with plausible confounders stated.
HypothesisA possible explanation that remains testable and uncertain.
ForecastAn estimate issued at a known time, with scope and confidence.
Confirmed outcomeA recorded result with a defined measurement and completeness level.

Record before interpretation

A useful evidence package identifies the garden context, date range, relevant plants or varieties, available photographs, weather and environmental context, grower actions, observed changes, harvest outcome, current interpretation, alternative explanations, confidence, and what remains unknown.

The goal is not to sound certain. The goal is to make the available evidence useful without hiding its boundaries.

Coverage and missing data

A photo shows only what is visible from a particular view at a particular time. Leaves can hide fruit. A missed capture can conceal a change. An unrecorded harvest can make a forecast look wrong or right for the wrong reason. GardenSeer should surface these limits alongside any interpretation.

Privacy-safe publication

Private captures are never publication-ready by default. A public image requires a separate derivative and review.

  1. Select the private source by an internal identifier.
  2. Create a separate derivative for a specific, reviewed use.
  3. Strip location, device, thumbnail, and unnecessary metadata.
  4. Inspect the full frame for people, addresses, screens, reflections, landmarks, and property detail.
  5. Publish only the approved derivative with factual alt text, context, rights, and allowed uses.

Current publication boundary: This site contains no raw captures, private garden records, customer data, automatic media feeds, or access to the authenticated application.

Corrections and learning

A later outcome can change the interpretation of an earlier record. Published work should be corrected, annotated, replaced, or withdrawn when better evidence requires it. Learning from a failed hypothesis is part of the method.