Reference for all available metrics, dimensions, aggregation functions, and filter operators in the dashboard query engine.

Metrics Reference

Dashboard widgets query one of five data views. Each view exposes a set of metrics you can aggregate and dimensions you can group by or filter on.

Data Views

Traces

One row per trace. Use this view for request-level monitoring.

Metrics

Metric	Description
`count`	Number of traces
`latency`	End-to-end trace duration (ms)
`totalTokens`	Sum of input + output tokens across all observations in the trace
`totalCost`	Estimated cost (USD) across all observations
`uniqueUserCount`	Distinct user IDs
`observationsCount`	Number of observations (spans/generations) per trace
`scoresCount`	Number of evaluation scores attached

Dimensions

Dimension	Description
`name`	Trace name set by the SDK
`userId`	User identifier passed at ingestion
`sessionId`	Session identifier
`environment`	Environment tag (e.g. `production`, `staging`)
`release`	Release version string
`version`	Application version
`tags`	Free-form tags array

Observations

One row per span, generation, or event. Use this view for model-level and operation-level analysis.

Metrics

Metric	Description
`count`	Number of observations
`latency`	Observation duration (ms)
`streamingLatency`	Streaming-specific latency (ms)
`timeToFirstToken`	Time from request to first token (ms)
`tokensPerSecond`	Overall throughput (tokens/s)
`outputTokensPerSecond`	Output-only throughput (tokens/s)
`inputTokens`	Input token count
`outputTokens`	Output token count
`totalTokens`	Input + output tokens
`inputCost`	Estimated input cost (USD)
`outputCost`	Estimated output cost (USD)
`totalCost`	Total estimated cost (USD)
`countScores`	Number of scores attached

Dimensions

Dimension	Description
`name`	Observation name
`type`	Observation type (`GENERATION`, `SPAN`, `EVENT`)
`level`	Log level (`DEFAULT`, `DEBUG`, `WARNING`, `ERROR`)
`providedModelName`	Model name as provided by the SDK
`promptName`	Name of the prompt template used
`environment`	Environment tag

Sessions

One row per session (a group of related traces). Use this view for multi-turn conversation analytics.

Metrics

Metric	Description
`count`	Number of sessions
`traceCount`	Average number of traces per session
`latency`	Session duration (ms)
`totalTokens`	Total tokens across all traces in the session
`totalCost`	Total cost (USD) across the session

Dimensions

Dimension	Description
`sessionId`	Session identifier
`userId`	User identifier
`environment`	Environment tag
`release`	Release version string
`version`	Application version
`tags`	Free-form tags

Scores (Numeric)

One row per numeric evaluation score. Use this view to track evaluation quality over time.

Metrics

Metric	Description
`count`	Number of scores
`avg`	Mean score value
`min`	Minimum score value
`max`	Maximum score value
`p50`	Median score
`p75`	75th percentile
`p90`	90th percentile
`p95`	95th percentile
`p99`	99th percentile

Dimensions

Dimension	Description
`name`	Score name (evaluator name)
`source`	Score source (`API`, `EVAL`, `HUMAN`)
`dataType`	Data type (`NUMERIC`, `BOOLEAN`)
`environment`	Environment tag

Scores (Categorical)

One row per categorical evaluation score.

Metrics

Metric	Description
`count`	Number of scores

Dimensions

Same as Scores (Numeric): name, source, dataType, environment.

Aggregation Functions

Function	Description
`count`	Row count
`sum`	Sum of values
`avg`	Arithmetic mean
`unique`	Count of distinct values
`min`	Minimum value
`max`	Maximum value
`p50`	50th percentile (median)
`p75`	75th percentile
`p90`	90th percentile
`p95`	95th percentile
`p99`	99th percentile

Percentile aggregations (p50–p99) are most meaningful for latency and cost metrics. They are computed using ClickHouse's quantile function.

Filters

You can add one or more filter conditions to any widget to restrict which rows are included in the query.

Filter Operators

Operator	Applies to	Description
`equals`	String, number	Exact match
`not equals`	String, number	Exclude exact match
`contains`	String	Substring match
`not contains`	String	Exclude substring match
`starts with`	String	Prefix match
`ends with`	String	Suffix match
`greater than`	Number	Numeric comparison
`less than`	Number	Numeric comparison
`is null`	Any	Field is absent/null
`is not null`	Any	Field is present

Multiple filters on the same widget are combined with AND logic.

Time Range and Granularity

All widgets operate on a time range (start date to end date). You can set a date preset per widget:

Preset	Window
Last 1 hour	1 h
Last 24 hours	24 h
Last 7 days	7 d (default)
Last 30 days	30 d
Last 90 days	90 d

For time-series widgets, granularity controls how data is bucketed along the time axis:

Granularity	Bucket size
`auto`	Chosen automatically based on date range
`hour`	1 hour buckets
`day`	1 day buckets
`week`	7 day buckets
`month`	Calendar month buckets

Metrics

On this page