Internal Developer Platform  ·  View 10 of 21  ·  Data

Data Flow

From declaration to projection, and the direction data is not allowed to travel.

Editable source SVG draw.io All views
Declared Catalog Descriptor team repo Deploy and Infra Intent Policy Source Ingested Descriptor Ingester webhook driven Bundle Builder signs · publishes Acted on Workload Reconciler Infra Reconciler Admission Evaluator Observed Runtime State AKS fleet Resource Graph Azure Repo Activity commits · PRs Projected Software Catalog PostgreSQL Dependency Graph Event and Metric Facts Data Explorer Consumed Component Page DORA and Adoption Control Coverage for audit owner lookup cost · deploys Internal Developer Platform — Data Flow External / third party Security / platform Application we own Data store synchronous batch Nothing flows the other way into the declared column: the platform never writes to a team's repository except as a pull request a human merges. v 1.0 · owner Platform Architecture · date 2026-09

Decisions

  • Nothing flows back into the declared column except as a pull request a human merges. The platform never silently edits a team's repository (ADR-01, ADR-03).
  • Observation is a first-class column: the platform reads the runtime and the resource graph rather than trusting its own record of what it created (ADR-09).
  • Control coverage is a consumer, which is how an off-road component still appears in the audit story (ADR-05).

Assumptions

  • Catalog freshness ≤ 10 minutes p95 from a deploy or ownership change (stated assumption).
  • Descriptor ingestion is webhook-driven with a periodic full reconcile as the backstop.

Risks

  • The projection can lag or be wrong. Divergence is reported in both directions rather than silently resolved, because a catalog that quietly corrects itself teaches nobody anything.
  • Owner lookup on the reconciliation path is the one place the catalog is in a critical path; it is cached and fails to last-known-good.