EventTracking

The plumbing that captures events reliably: client SDKs, event schema, batching, delivery guarantees.

Why we need this / what value this brings

Bad event plumbing (dropped events, no schema) silently corrupts every downstream analysis built on top of it.

When to use this

As soon as any analytics is being collected at all — the plumbing has to be trustworthy from day one.

How to use or implement this

Pick a delivery guarantee (at-least-once is usually fine with idempotent processing downstream), version your event schema, and validate events at write time.

Research questions

• At-least-once vs exactly-once event delivery — does double-counting a booking event matter?
• How do you version an event schema without breaking historical analysis?

---

Empty folder — drop notes, links, and findings here as you research.