🪢 Open source agent evals & observability: Trace, evaluate, and improve LLM applications with one open platform.
TypeScript analyticsautogenevaluationlangchainlarge-language-modelsllama-indexllmllm-evaluationllm-observabilityllmopsmonitoringobservabilityopen-sourceopenaiplaygroundprompt-engineeringprompt-managementself-hostedycombinator
/**
* Shared ClickHouse query building utilities for score analytics
*
* These helpers are used by both buildEstimateQuery and buildScoreComparisonQuery
* to construct SQL fragments for filtering and sampling.
*/
/**
* Build object type filter for score queries
*
* Generates SQL WHERE clause fragment to filter scores by attachment type.
* Ensures exclusive filtering (exactly one non-NULL ID) for trace, observation,
* session, and dataset_run types.
*
* @param objectType - The object type to filter by
* @returns SQL WHERE clause fragment for object type filtering
*
* @example
* ```typescript
* buildObjectTypeFilter("trace")
* // Returns: "AND trace_id IS NOT NULL AND observation_id IS NULL AND session_id IS NULL AND dataset_run_id IS NULL"
*
* buildObjectTypeFilter("all")
* // Returns: ""
* ```
*/
export function buildObjectTypeFilter(objectType: string): string {
if (objectType === "all") return "";
if (objectType === "trace")
return "AND trace_id IS NOT NULL AND observation_id IS NULL AND session_id IS NULL AND dataset_run_id IS NULL";
if (objectType === "observation") return "AND observation_id IS NOT NULL";
if (objectType === "session")
return "AND session_id IS NOT NULL AND observation_id IS NULL AND trace_id IS NULL AND dataset_run_id IS NULL";
if (objectType === "dataset_run")
return "AND dataset_run_id IS NOT NULL AND trace_id IS NULL AND observation_id IS NULL AND session_id IS NULL";
return "";
}
/**
* Build sampling expression for hash-based sampling
*
* Uses cityHash64 on composite key (trace_id, observation_id, session_id, dataset_run_id)
* to ensure deterministic pseudo-random sampling that preserves matched pairs across
* score1 and score2 queries.
*
* The cityHash64 function provides uniform distribution, and modulo operation ensures
* consistent sampling across different score tables.
*
* @param samplingPercent - Percentage to sample (0-100)
* @returns SQL expression for hash-based sampling
*
* @example
* ```typescript
* buildSamplingExpression(10)
* // Returns: "cityHash64(...) % 100 < 10"
*
* buildSamplingExpression(100)
* // Returns: "cityHash64(...) % 100 < 100" (no sampling, all data)
* ```
*/
export function buildSamplingExpression(samplingPercent: number): string {
return `cityHash64(
coalesce(trace_id, ''),
coalesce(observation_id, ''),
coalesce(session_id, ''),
coalesce(dataset_run_id, '')
) % 100 < ${samplingPercent}`;
}