Audit Log Service · View 13 of 40 · 3 · Structure
Question 1 answered
- Per-tenant chains, anchored into a global checkpoint every 15 minutes. Not one global chain and not independent per-tenant chains.
- One global chain gives the strongest ordering claim, but it serialises sealing at 100 k events/s and leaks cross-tenant volume through every Merkle path an export must reveal.
- Independent per-tenant chains parallelise cleanly but leave an attacker with tenant-scoped access a small structure to forge. The global checkpoint binds every tenant head, so forging one tenant's history means forging a signature over all of them.
What it buys
- Inclusion proofs are tenant-blind: a proof path stays inside its own tenant's tree and reveals nothing about anyone else's volume or timing.
- 500 chains extend in parallel, which removes sealing throughput as a design constraint.
- Cross-tenant ordering is still evidence when it needs to be, because the checkpoint fixes every tenant head at the same instant against an external timestamp.
Risks
- Between checkpoints, a tenant's chain is only as strong as the chain itself. The exposure window is 15 minutes, and shortening it is a cost decision (HSM operations and ledger writes), not a redesign.