Principle:Truera Trulens Feedback Result Retrieval
| Knowledge Sources | |
|---|---|
| Domains | LLM_Evaluation, Data_Analysis |
| Last Updated | 2026-02-14 08:00 GMT |
Overview
A result collection pattern that retrieves and consolidates asynchronous feedback evaluation scores into a structured DataFrame for analysis.
Description
Feedback Result Retrieval bridges the gap between asynchronous feedback evaluation and synchronous analysis. After application recording, feedback functions run asynchronously (either in the same process or server-side). This principle provides a blocking retrieval mechanism that:
- Waits for all records to be persisted to the database
- Triggers immediate computation of pending feedback evaluations
- Waits for all feedback results to be available
- Returns a structured pandas DataFrame with records as rows and feedback scores as columns
This is essential for programmatic analysis workflows where you need to inspect evaluation results immediately after recording, rather than waiting for the dashboard.
Usage
Use this principle after recording application traces when you need immediate programmatic access to feedback scores. Call it on a Recording object after exiting the context manager. Set an appropriate timeout for long-running evaluations.
Theoretical Basis
This pattern implements the Future/Promise resolution pattern — feedback evaluations are submitted as asynchronous tasks, and retrieval blocks until all promises are resolved or a timeout is reached.
Pseudo-code Logic:
# Abstract result retrieval pattern
with app as recording:
app.invoke(query)
# Block until results are available
results_df = recording.retrieve_feedback_results(timeout=180)
# results_df: DataFrame with record_id index, feedback name columns, score values