Cost Allocation & Showback Platform  ·  View 19 of 22  ·  6 · Operations

Observability

Five signal classes across five stages. Only five cells page a human, and four of them are correctness rather than availability.

Editable source SVG draw.io All views
Ingest Conform Allocate Publish Serve Freshness Export landed by 06:00 Conformed by 09:00 Facts by 10:00 T+1 Frozen in 5 days Mart lag under 1 h Correctness Row count vs manifest Invoice variance pages Digest reproducibility pages Sum of lines = total pages Drill-through sums back Coverage Providers reporting Unknown SKUs Unallocated % SLO 2% Teams with a statement Volume & cost Bytes landed / day Rows conformed Run minutes & spot loss Statements per close Query cost per team Audit Credential use Quarantine events Policy version applied Freeze & restate events Cross-team reads Observability — What Is Watched at Each Stage, and What Pages Only five cells page a human, and four of them are correctness rather than availability. A dashboard that is an hour stale is an inconvenience; a number that does not reconcile is the failure this platform exists to avoid. v 1.0 · owner Platform Architecture · date 2026-09

What pages

  • Invoice variance outside tolerance
  • An allocation run whose digest does not reproduce
  • A statement whose lines do not sum to its total
  • Unallocated share breaching the 2% SLO

What does not page

  • Dashboard staleness under an hour — an inconvenience, not a failure
  • A single missed export inside the T+2 cutoff
  • Spot interruption on the batch tier

The principle

  • For a measurement platform, being wrong is worse than being late. The alerting reflects that and nothing else
◀  Policy Promotion Pipeline All views The Monthly Close Loop  ▶
Solution Architecture v1.0 · Microsoft Azure · Cost & Efficiency Architecture · 2026-0919-observability