Datasets:
roanbrasil
/
devops-gitops-corpus

Dataset card Data Studio Files
xet
 Community
1
text
stringlengths
0
59.1k
{ type: "meanScore", min: 0.8 },
{ type: "passRate", min: 0.9, scorerId: "exactMatch" },
];
```
All criteria must pass for the run to succeed. Criteria marked `severity: "warn"` don't fail the run but are reported in the summary.
#### `label` and `description`
Human-readable strings for dashboards and logs:
```ts
label: "Nightly Regression Suite",
description: "Validates prompt changes against production scenarios."
```
#### `tags`
Array of strings attached to the run for filtering and search:
```ts
tags: ["nightly", "production", "v2-prompts"];
```
#### `metadata`
Arbitrary key-value data included in the run result:
```ts
metadata: {
branch: "feature/new-prompts",
commit: "abc123",
environment: "staging",
}
```
#### `experiment`
Binds the run to a named experiment in VoltOps:
```ts
experiment: {
name: "support-regression",
id: "exp-123", // optional - explicit experiment ID
autoCreate: true, // optional - create experiment if missing (default: true)
}
```
When `autoCreate` is true and the experiment doesn't exist, VoltOps creates it on first run.
#### `voltOps`
VoltOps integration settings:
```ts
voltOps: {
client: voltOpsClient, // optional - SDK instance
triggerSource: "ci", // optional - "ci", "manual", "scheduled", etc.
autoCreateRun: true, // optional - defaults to true
autoCreateScorers: true, // optional - register scorers in VoltOps
tags: ["regression", "v2"], // optional - additional tags
}
```
## Dataset Items
Each item in the dataset has this structure:
```ts
interface ExperimentDatasetItem {
id: string; // unique identifier
input: unknown; // passed to runner
expected?: unknown; // passed to scorers
label?: string | null; // human-readable description
extra?: Record<string, unknown> | null; // additional metadata
datasetId?: string; // VoltOps dataset ID (auto-populated)
datasetVersionId?: string; // VoltOps version ID (auto-populated)
datasetName?: string; // Dataset name (auto-populated)
metadata?: Record<string, unknown> | null; // item-level metadata
}
```
The `input` and `expected` types are generic - use any structure your runner and scorers expect.
## Scorers
Scorers compare runner output to expected values or apply custom validation. Each scorer returns a result with:
- `status` - `"success"`, `"error"`, or `"skipped"`
- `score` - Numeric value (0.0 to 1.0 for normalized scorers)
- `metadata` - Additional context (e.g., token counts, similarity details)
- `reason` - Explanation for the score (especially for LLM judges)
- `error` - Error message if status is `"error"`
### Heuristic Scorers (No LLM Required)
```ts
import { scorers } from "@voltagent/scorers";
// String comparison