Recipe catalog / grounding-level
Grade draft grounding
How much of the substantive content in draft is backed by evidence, on a five-level rubric?
You need one graded measure of how well a generated answer sticks to its retrieved sources before sending or ranking it.
Explore this recipe interactively ยท Source and implementation guide
Use grounding-level in TypeScript
Install with npm install jev-recipes. Requires Node.js 22.9 or newer and ES modules. Set TYPESAFE_API_KEY in your server environment for live calls, which send input to TypeSafe and use API quota. See the installation guide.
import { groundingLevel } from 'jev-recipes/grounding-level';
const result = await groundingLevel({
"draft": "Your Starter plan includes 5 GB of storage and unlimited collaborators. Files larger than 2 GB must be uploaded through the desktop app, and deleted files can be restored for 30 days.",
"evidence": "Starter plan: 5 GB storage, up to 10 collaborators. Uploads larger than 2 GB require the desktop application. Deleted files can be restored for 30 days.",
"minConfidence": 0.8
});
console.log(result);
Input contract
| Field | Type | Needed |
|---|---|---|
| draft | string | Required |
| evidence | string | Required |
| minConfidence | number | Optional |
Full input and result schemas
{
"input": {
"$schema": "https://json-schema.org/draft/2020-12/schema",
"type": "object",
"properties": {
"draft": {
"type": "string"
},
"evidence": {
"type": "string"
},
"minConfidence": {
"type": "number",
"minimum": 0,
"maximum": 1
}
},
"required": [
"draft",
"evidence"
]
},
"result": {
"$schema": "https://json-schema.org/draft/2020-12/schema",
"type": "object",
"properties": {
"model": {
"type": "string"
},
"usage": {
"type": "object",
"properties": {
"input_tokens": {
"type": "integer",
"minimum": 0,
"maximum": 9007199254740991
},
"output_tokens": {
"type": "integer",
"minimum": 0,
"maximum": 9007199254740991
}
},
"required": [
"input_tokens",
"output_tokens"
],
"additionalProperties": false
},
"status": {
"type": "string",
"enum": [
"ready",
"review"
]
},
"score": {
"type": "number",
"minimum": 0
},
"level": {
"type": "integer",
"minimum": 0,
"maximum": 9007199254740991
},
"confidence": {
"type": "number",
"minimum": 0,
"maximum": 1
},
"probabilities": {
"type": "object",
"propertyNames": {
"type": "string"
},
"additionalProperties": {
"type": "number",
"minimum": 0,
"maximum": 1
}
},
"grounding": {
"type": "string",
"enum": [
"none",
"weak",
"partial",
"mostly",
"full"
]
}
},
"required": [
"model",
"usage",
"status",
"score",
"level",
"confidence",
"probabilities",
"grounding"
],
"additionalProperties": false
}
}Saved example result
This hand-authored response demonstrates the contract. It is not a model accuracy measurement. Run it without an API key: npx jev-recipes demo grounding-level.
{
"model": "demo-fixture",
"usage": {
"input_tokens": 0,
"output_tokens": 0
},
"status": "ready",
"score": 2.91,
"level": 3,
"confidence": 0.85,
"probabilities": {
"0": 0,
"1": 0.02,
"2": 0.09,
"3": 0.85,
"4": 0.04
},
"grounding": "mostly"
}
Evaluation evidence
No verified live accuracy measurement is available. Evaluate representative cases before using this decision in your workflow.
Use the evaluation guide to measure this decision on your own labeled cases.
Limitations
- Reports how much is grounded, not which statements are unsupported. Use verify per claim to locate them.
- Grades support from the supplied evidence only. A true statement from outside knowledge still counts as unsupported.
Related recipes
- verify: Use verify for a supported, contradicted, or unsupported verdict on a single claim.
- answer-relevance: Use answer-relevance to check that the draft addresses the question, not just the sources.