Defining metrics

Design the numbers your project will prove: totals, averages, and percentages

Metrics are the backbone of a Mirage project. They define what the AI looks for in field reports, what stewards approve, and what donors ultimately see. Take a few minutes to design them well: everything downstream inherits their quality.

​

Anatomy of a metric

Every metric has:

• Label: the donor-facing name (“Wells dug”, “Lives impacted”, “Water access rate”).
• Description: what it measures and why it matters. This also helps the AI extract the right values from reports.
• Aggregation: how individual report values roll up into one number.
• Presentation: how the final number is formatted for display.

​

Aggregation types

​

Scalar metrics

One value per report, combined across reports:

Aggregator	Use it for	Example
Sum	Cumulative totals	Total wells dug across all reports
Average	Typical values	Average satisfaction score (1–5)
Count	How many reports meet a condition	Number of site visits completed
Min / Max	Extremes and thresholds	Deepest well drilled

Scalar aggregation can be confidence-weighted, so values the AI extracted with high confidence count more than uncertain ones.

​

Ratio metrics

Percentages computed across reports:

Aggregator	Use it for	Example
Boolean share	”What fraction said yes?”	% of households with clean water access
Filtered ratio	A numerator over a denominator	Graduation rate: students passing ÷ students enrolled

​

Presentation options

Choose how the aggregated number reads on dashboards and donor reports:

• Number: with configurable decimals and compact notation (1.2K, 3M).
• Percent: with optional underlying counts shown alongside.
• Currency: symbol, placement, and decimals configurable.
• Threshold: display “Met goal” / “Below target” based on a target value, for metrics that are really pass/fail commitments.

​

How values get into metrics

You never type metric values by hand. When a field report is finalized, Mirage’s AI reads the answers, photos, transcriptions, and media summaries, and proposes contributions: candidate values for your metrics, each with a confidence score and a link to its source. Stewards approve or reject each contribution in the review queue; only approved contributions feed the aggregate.

Write question prompts with your metrics in mind. If your metric is “households served”, make sure a reporting task asks a question whose answer contains that number. The AI extracts values from what reporters actually submit. See Reporting tasks.

Metrics can be edited from project settings, but changing aggregation mid-project changes what your historical numbers mean. Prefer adding a new metric over redefining one that already has approved contributions.