Confidence and Reliability in Plant Diagnosis
This article includes historical API examples. Read the current API documentation for supported plants and fields.
Confidence and reliability answer different questions
A diagnosis can have high confidence and still be wrong. A photo can hide symptoms, show several problems, or resemble more than one condition.
Per-condition confidence describes the strength of a particular prediction. reliability_score is a separate estimate of the reliability of a plant result. Neither value confirms the cause of a plant problem.
Read the scores within their scope
| Field | Meaning | Scope |
|---|---|---|
confidence | Strength of a condition or pest prediction | Each returned condition or pest |
reliability_score | Estimate of result reliability | Optional on some unhealthy cannabis results |
reliability_score ranges from 0 to 1. A higher score indicates higher estimated reliability. It does not measure a proven chance of correctness for your photo.
The current tomato response does not include reliability_score. An unhealthy cannabis result can also omit the field. Treat absence as “no score available,” rather than zero or permission to act.
Use the result to guide review
An integration can log the result, request another photo, or flag a plant for inspection. Read any available scores alongside the visible symptoms and other possible causes.
A score threshold is a rule your application chooses. The API does not provide a universal threshold for safe automatic treatment. Test any rule on your own images before using it, and consider the effect of a wrong result.
A high score alone does not justify changing irrigation, adding nutrients, or applying a treatment. Inspect the plant before choosing an action.
The historical field change
PlantLab replaced diagnostic_confidence and safety_classification with reliability_score in the historical schema change linked below. The fields describe different signals. Do not treat the new field as a direct numerical replacement for either old field.
For a current integration, use the current contract and handle the field as optional. The historical migration article retains the dated examples.
FAQ
Does high confidence mean the diagnosis is correct?
No. Confidence expresses the strength of the prediction. A strong prediction can be wrong.
Does reliability guarantee correctness?
No. It is an estimate that can help route a result for closer review. It differs from condition confidence.
What if reliability is missing?
Treat it as unavailable. Do not turn a missing score into zero or a high score. Check the plant and the other response fields.
What threshold permits automatic treatment?
This article does not establish one. Test your own review rules and inspect the plant before treatment.
PlantLab is currently free to use at plantlab.ai. Read the API documentation for current plants, fields, and account limits.
Related reading: