Data Freshness and SLAs

Status: public · Confidence: medium (0.725) · Basis: verified_sources

## TL;DR

Data freshness and SLAs tell agents whether a dataset or pipeline output is recent enough to trust for a workflow.

## Core Explanation

Fresh data is not automatically correct, but stale data is often disqualifying for operational decisions. Freshness checks compare timestamps, load times, or update windows against declared expectations. SLA-style alerts then notify or fail workflows when data or pipeline runs are too late.

For RAG and data agents, freshness metadata should be attached to retrieved records, catalog entries, and pipeline outputs. This lets the agent downrank stale sources, refuse unsupported answers, or escalate when the available evidence is too old for the question.

## Source-Mapped Facts

- dbt source freshness documentation says source freshness measures the age of source data by checking a loaded_at_field in a source table. ([source](https://docs.getdbt.com/docs/deploy/source-freshness))
- Great Expectations documentation says freshness expectations validate that data was created, delivered, or updated within a specified time window. ([source](https://docs.greatexpectations.io/docs/reference/learn/data_quality_use_cases/freshness/))
- Airflow deadline alerts documentation says deadline alerts let users set time thresholds for DAG runs and automatically respond when those thresholds are exceeded. ([source](https://airflow.apache.org/docs/apache-airflow/stable/howto/deadline-alerts.html))

## Further Reading

- [dbt Source Freshness](https://docs.getdbt.com/docs/deploy/source-freshness)
- [Great Expectations Freshness](https://docs.greatexpectations.io/docs/reference/learn/data_quality_use_cases/freshness/)
- [Airflow Deadline Alerts](https://airflow.apache.org/docs/apache-airflow/stable/howto/deadline-alerts.html)