Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.galtea.ai/llms.txt

Use this file to discover all available pages before exploring further.

The Knowledge Retention metric is one of several non-deterministic Metric Galtea uses to evaluate your LLM-based chatbot’s ability to retain and consistently apply factual information shared earlier in a conversation. It analyzes the entire conversational history to determine whether the model recalls and reuses relevant facts when generating new responses. This is particularly useful for long, multi-turn dialogues where context accumulation and memory play a crucial role in the user experience.

​
Evaluation Parameters

To compute the knowledge_retention metric, the following parameters are required in every turn of the conversation:
  • input: The user message in the conversation.
  • actual_output: The LLM-generated response to the user message.
This metric will evaluate the whole conversation, including all turns, to simulate a memory-check process across multiple turns.

​
How Is It Calculated?

The knowledge_retention score is computed using an LLM-as-a-judge approach:
  1. Identify Knowledge Anchors: The LLM scans user inputs to identify specific facts, preferences, constraints, or context (e.g., names, locations, specific numbers).
  2. Verify Recall: The LLM checks if the agent recalled and applied this information in subsequent turns.
  3. Check Consistency: The LLM evaluates whether the agent contradicted previously established information, asked for information already provided, or ignored constraints set earlier.
The metric assigns a binary score:
  • Score 1.0 (Good Retention): The agent correctly recalled relevant information or no specific memory recall was required (and no errors were made).
  • Score 0.0 (Poor Retention): The agent forgot information, contradicted itself, or asked redundant questions about known facts.

​
Suggested Test Case Types

The Knowledge Retention metric is effective for evaluating Behavior test cases in Galtea, particularly:
  • Long multi-turn conversations where the user shares preferences, constraints, or facts early on.
  • Personalized assistant scenarios where the agent must recall user-provided details.
  • Complex workflows where information from one step is needed in a later step.