Consent & Privacy Service  ·  View 11 of 22  ·  Data

State Classes

Three classes ordered by what happens if the data is lost. Only the top row has an RPO worth arguing about.

Editable source SVG draw.io All views
Irreplaceable — RPO 0, never rebuilt Consent ledger Decision entries 7 yr, append-only Evidence Audit records write-once Case outcomes 3 yr Keys Per-subject data keys loss = erasure Reconstructible — rebuilt from the ledger at 20x real time Derived reads Current state RPO 60 s Batch snapshot Enforcement caches RPO n/a Operationally durable — losing it reopens work, not truth In flight Open case state resumable Identity index re-derivable, slowly Suppression list rebuilt from cases rebuild on completion State Classes — Ordered by What Happens If It Is Lost Data store Security / platform Application we own batch event / async Only the top row has an RPO worth arguing about. Everything in the middle row is deliberately disposable, which is what makes a bad deploy survivable. v 1.0 · owner Security & Identity Architecture

Decisions

  • The ledger, the audit store and the per-subject keys are irreplaceable: RPO 0, committed to a regional quorum before a capture is acknowledged.
  • Projections, snapshots and enforcement caches are deliberately disposable. That is what turns a corrupt projection or a bad decision deploy into a rebuild rather than an incident with no exit.
  • Losing a per-subject key is indistinguishable from erasing that subject. The key store is therefore in the irreplaceable row and backed up with the ledger, not with the projections.

Assumptions

  • RPO 0 for ledger, audit and keys; RPO 60 s for derived projections; RTO 10 minutes for a region's decision path, 60 minutes for its full projection rebuild.
  • Ledger 40 TB over 7 years after compression; hot for recent entries, archive beyond, retrievable inside the regulator-response window rather than instantly.

Risks

  • The identity index is marked re-derivable "slowly", which is optimistic: re-deriving it depends on forty systems still holding the linkage they held when it was built. Treat it as closer to irreplaceable than the row suggests.
  • A 20× replay rate is an assumption, not a measurement. The whole 60-minute RTO rests on it, so it is one of the three things the prototype must falsify.