Groundedness
Checks whether a response is strictly grounded in the provided context, flagging any information the context doesn't support.
Groundedness checks whether a response is strictly based on the provided context, with no outside information introduced. Run it wherever an answer must be traceable back to a source.
What it does
Groundedness is an LLM-as-Judge eval. It reads the context and the generated output (and optionally the input), then returns a pass/fail on whether the response is fully supported by that context.
Input
| Required Input | Type | Description |
|---|---|---|
output | string | The output generated by the model |
context | string | The context provided to the model |
| Optional Input | Type | Description |
|---|---|---|
input | string | The input provided to the model |
Output
| Field | Type | Description |
|---|---|---|
| Result | Pass / Fail | Passed means the response is fully grounded in the provided context, Failed means the response introduces unsupported information |
| Reason | string | A plain-language explanation of the groundedness assessment |
Run it from code
Call evaluate() with the template name and the eval’s required inputs. It returns the score and the reason.
Note
Before running: install the SDK and set FI_API_KEY / FI_SECRET_KEY. The model argument in the snippets is the evaluator model Future AGI uses to run the eval; turing_flash is a fast default.
from fi.evals import evaluate
result = evaluate(
"groundedness",
input="The Earth orbits around the Sun in how many days?",
context="The Earth completes one orbit around the Sun every 365.25 days",
output="365.25 days",
model="turing_flash",
)
print(result.score)
print(result.reason)import { evaluate } from "@future-agi/ai-evaluation";
const result = await evaluate(
"groundedness",
{
input: "The Earth orbits around the Sun in how many days?",
context: "The Earth completes one orbit around the Sun every 365.25 days",
output: "365.25 days"
},
{ modelName: "turing_flash" }
);
console.log(result);When to use
Run Groundedness wherever an answer needs to be traceable to a source document and any unsupported addition is a problem.
- Text, audio, and chat outputs generated from a fixed context
- RAG and retrieval pipelines, to confirm answers don’t extend past the retrieved material
- Hallucination checks, alongside Context Adherence and Detect Hallucination, for a stricter Pass/Fail read on grounding
What to do when Groundedness fails
Reassess the provided context for completeness and clarity, ensuring it includes all necessary information to support the response.
Examine the response for any elements not supported by the context, and adjust it to improve alignment with the given information.
Questions & Discussion