1 min lesson
Data observability & quality
Show why this lesson detail matters: "The job description names it directly: build observability, alerting, SLAs/SLOs and operational standards across the platform."
Step 1 of 3
The job description names it directly: build observability, alerting, SLAs/SLOs and operational standards across the platform. In the loop that line becomes a probe - when a column goes null at 3am, do you find out from a dashboard or from an analyst's angry Slack?
Cursor ingests billions of product-telemetry events a day and a slice of that feeds agent and model work. Bad data there is not a cosmetic bug. A silently broken bronze table can poison a gold metric a VP reads or skew a feature an agent learns from. So the interviewer wants a mental model of what can go wrong, not a list of vendor names.
Learn more
Advanced table
Freshness and volume catch the obvious outages
- Dimension
- Freshness
- Question it answers
- Did new data land on time?
- Concrete check at Cursor scale
- Bronze telemetry partition for the current hour exists and is < 15 min late
- Dimension
- Volume
- Question it answers
- Did the right amount arrive?
- Concrete check at Cursor scale
- Event count within ±20% of the same hour last week; zero rows = page
- Dimension
- Schema
- Question it answers
- Did the shape change underneath us?
- Concrete check at Cursor scale
- New/dropped columns, type drift, an enum that grew a value the parser ignores
- Dimension
- Distribution
- Question it answers
- Are the values plausible?
- Concrete check at Cursor scale
- Null rate on
user_id, p99 oflatency_ms, ratio of a category that suddenly flips
- Dimension
- Lineage
- Question it answers
- What breaks if this breaks?
- Concrete check at Cursor scale
- Map from a bronze asset to every silver/gold/serving consumer downstream
| Dimension | Question it answers | Concrete check at Cursor scale |
|---|---|---|
| Freshness | Did new data land on time? | Bronze telemetry partition for the current hour exists and is < 15 min late |
| Volume | Did the right amount arrive? | Event count within ±20% of the same hour last week; zero rows = page |
| Schema | Did the shape change underneath us? | New/dropped columns, type drift, an enum that grew a value the parser ignores |
| Distribution | Are the values plausible? | Null rate on user_id, p99 of latency_ms, ratio of a category that suddenly flips |
| Lineage | What breaks if this breaks? | Map from a bronze asset to every silver/gold/serving consumer downstream |
Freshness and volume catch the obvious outages; schema and distribution catch the quiet corruptions that hurt most.