Datasets:
roanbrasil
/
devops-gitops-corpus

Dataset card Data Studio Files
xet
 Community
1
text
stringlengths
0
59.1k
environment: "prod-us-east";
environment: "local-dev";
```
#### `sampling`
Controls what percentage of interactions are scored. Use sampling to reduce latency and LLM costs on high-volume agents.
**Ratio-based:**
```ts
sampling: {
type: "ratio",
rate: 0.1, // score 10% of interactions
}
```
**Count-based:**
```ts
sampling: {
type: "count",
rate: 100, // score every 100th interaction
}
```
**Always sample:**
```ts
sampling: { type: "ratio", rate: 1 } // 100%
```
When unspecified, sampling defaults to scoring every interaction (`rate: 1`).
Sampling decisions are made independently for each scorer. Set sampling at the eval level (applies to all scorers) or per-scorer to override.
#### `scorers`
Map of scorer configurations. Each key identifies a scorer instance, and the value defines the scorer function and parameters.
```ts
scorers: {
moderation: {
scorer: createModerationScorer({ model, threshold: 0.5 }),
},
keyword: {
scorer: keywordMatchScorer,
params: { keyword: "refund" },
},
}
```
#### `redact`
Function to remove sensitive data from evaluation payloads before storage. Called synchronously before scoring.
```ts
redact: (payload) => ({
...payload,
input: payload.input?.replace(/\b\d{4}-\d{4}-\d{4}-\d{4}\b/g, "[CARD]"),
output: payload.output?.replace(/\b\d{4}-\d{4}-\d{4}-\d{4}\b/g, "[CARD]"),
});
```
The redacted payload is stored in observability but scoring uses the original unredacted version.
## Scorer Configuration
Each entry in the `scorers` map has this structure:
```ts
{
scorer: LocalScorerDefinition | (() => Promise<LocalScorerDefinition>),
params?: Record<string, unknown> | ((payload: AgentEvalContext) => Record<string, unknown>),
sampling?: SamplingPolicy,
id?: string,
onResult?: (result: AgentEvalResult) => void | Promise<void>,
}
```
### Fields
#### `scorer` (required)
The scoring function. Use prebuilt scorers from `@voltagent/scorers` or custom implementations via `buildScorer`.
**Prebuilt scorer:**
```ts
import { createModerationScorer } from "@voltagent/scorers";
scorer: createModerationScorer({ model, threshold: 0.5 });
```
**Custom scorer:**
```ts
import { buildScorer } from "@voltagent/core";
const customScorer = buildScorer({