Skip to main content

Documentation Index

Fetch the complete documentation index at: https://www.mireye.ai/llms.txt

Use this file to discover all available pages before exploring further.

Mireye attaches a confidence rating to every field value it returns. The rating is a three-bucket string — HIGH, MEDIUM, or LOW — not a numeric score. A string is easier to interpret in dashboards, reports, and audit trails, and easier to act on without requiring statistical literacy. Every field in a /v1/fetch response and every citation in a /v1/ask response carries its own confidence rating, so you can evaluate data quality at the field level rather than relying on a single summary signal.

​
The three confidence levels

HIGH

Value fetched from the authoritative source with no known issues. The returned value matches the documented precision and coverage for that source. This is the expected level for most fields under normal conditions.

MEDIUM

Value fetched successfully, but from a fallback source rather than the primary one, or the primary source has partial coverage at this coordinate. The value is usable but carries additional uncertainty.

LOW

Value is present but derived indirectly or sourced from a lower-resolution dataset. For example, a coarser national raster used when a higher-resolution regional dataset has a gap. Treat LOW-confidence values with caution before using them in decisions.

​
Confidence on /v1/ask responses

/v1/ask responses include a top-level confidence field that summarizes across all citations used to construct the answer.
The overall confidence of an /v1/ask response reflects the weakest citation: if all citations are HIGH, the overall is HIGH; any MEDIUM or LOW citation pulls the overall down to match. If you need field-level detail, inspect the individual confidence value in each citation object rather than relying on the top-level summary.
For example, an answer that draws on three HIGH-confidence fields and one MEDIUM-confidence field returns "confidence": "MEDIUM" at the response level, even though most of the underlying data was high quality.

​
When confidence is UNKNOWN

If Mireye reached the upstream source but the source returned a semantically null value — meaning no data exists for that coordinate rather than a fetch failure — the field value is null and confidence is UNKNOWN. This condition surfaces in the field’s notes field with a plain-language explanation. 
{
  "dominant_crop_5y": {
    "value": null,
    "unit": null,
    "source": "USDA_CDL",
    "source_url": "https://www.nass.usda.gov/Research_and_Science/Cropland/",
    "confidence": "UNKNOWN",
    "fetched_at": "2026-05-24T22:15:10.110Z",
    "ttl_seconds": 2592000,
    "notes": "Point is in a non-cultivated area; CDL reports no crop class."
  }
}
Semantic nulls are not the same as partial_failures. A semantic null means Mireye reached the source and the source confirmed there is no value for this location. A partial_failure means Mireye could not reach the source at all. Both conditions are always surfaced explicitly — Mireye never silently drops a field or substitutes a placeholder.

​
Using confidence in your application

How you act on confidence depends on your use case:
  • Regulatory and audit workflows: Filter to HIGH-confidence fields only before including values in a filing, report, or compliance record. If a required field does not return HIGH, flag it for manual review rather than proceeding with a lower-quality value.
  • Exploratory and screening workflows: MEDIUM confidence is generally acceptable when you are triaging a large portfolio or doing initial site selection. Use the results to narrow the candidate set, then pull HIGH-confidence data for the shortlisted sites.
  • Underwriting and lending decisions: Flag any LOW-confidence value for human review before it enters a pricing model or credit decision. Document the flag and the reviewer’s resolution in your audit trail. Do not substitute a LOW-confidence value for a missing HIGH-confidence one without explicit sign-off.
For batch workflows that process many coordinates, filter the response at read time using the confidence field on each value object. This lets you route HIGH values to automated pipelines and MEDIUM/LOW values to a review queue without making separate API calls.