Stable, Scoped Query Outputs

When writing KQL for Sentinel (parsers, analytic rules, hunts), make query outputs and intermediate datasets stable and correct:

• Prefer explicit project of the expected/normalized columns (avoid project-away) so newly added source columns won’t change the published schema.
• Apply required scoping and data-quality filters before arg_max/summarize (e.g., vendor scoping, excluding deleted/baseline-invalid records).
• Ensure deduplication via summarize on the intended entity key (e.g., SrcIpAddr) so you don’t emit duplicate entities.
• Guard against schema drift across connector versions by unioning V2/V3 tables when fields move.

Example patterns:

// 1) Stable output schema (no project-away)
SourceTable
| where DeviceVendor =~ "Ubiquiti"
| summarize count() by SrcIpAddr
| project  // explicitly list the final ASIM/expected columns
    TimeGenerated,
    EventType,
    SrcIpAddr,
    DeviceVendor;

// 2) Version drift handling
union isfuzzy=true
    SalesforceServiceCloudV2_CL,
    SalesforceServiceCloudV3_CL
| ...;

// 3) Baseline correctness before arg_max
BaselineTable
| where LifecycleStatus != "Deleted"
| summarize arg_max(Timestamp, *) by AgentId
| ...;

Adopting this standard reduces silent schema changes, incorrect baseline/detection logic, and duplicate or mis-resolved entities.