Add prediction-drift alerts on serving output

[gelluisaac]

Title

Add prediction-drift alerts on serving output

Context

AstroML is a research-driven Python framework for building dynamic graph machine learning models on the Stellar blockchain (ingestion -> graph -> train). This issue is a contributor-friendly task intended to be picked up by the community. Please ask for clarification in the discussion thread before starting, and reference this issue in your PR.

Task

Detect drift between training label distribution and serving predictions and alert when deviation exceeds thresholds.

Suggested files / areas

astroml/serving, astroml/observability

Acceptance Criteria

• Metrics/logs/traces are exposed in a standard format
• Alerts have clear thresholds and guidance
• Documented in the monitoring guide

Difficulty: Hard

This issue is ideal for a first-time contributor. If you would like to work on it, comment to claim it and maintainers will assign it to you.

[Classictowo]

@Classictowo has applied to work on this issue as part of the Stellar Wave Program's 8th wave.

Hello Maintainers, I have reviewed the issue and fully understand the requirements and acceptance criteria.I’m confident I can implement a solution that meets the expected design specifications.
Kindly assign this to me I’ll ensure timely and quality delivery. Thank you.

ℹ️ Repo Maintainers: To accept this application, review their application or assign @Classictowo to this issue.