Skip to content

Learning path: Product / PM

This page explains Learning path: Product / PM and how it fits into the RecSys suite.

What you care about

  • What artifacts exist and what they mean
  • How freshness works (daily windows)
  • How to roll back if something goes wrong
  • What "data quality" means in practice

Read in this order

1) start-here.md 1) explanation/artifacts-and-versioning.md 1) operations/slos-and-freshness.md 1) operations/runbooks/pipeline-failed.md

Key concepts

  • Artifacts are versioned and rollbackable because production needs safe

recovery.

  • Manifest pointers are updated last so serving never points to missing blobs.
  • Validation gates exist to prevent bad artifacts from reaching users.

Practical questions (and where answered)

  • "How often do recommendations update?"
  • operations/slos-and-freshness.md
  • "Can we revert to yesterday's artifact?"
  • how-to/rollback-manifest.md
  • "What if data is missing for a day?"
  • explanation/data-lifecycle.md