Persistence substrate
Dot.Memory is the knowledge-graph storage, vector indexes, and Knowledge Pack archive the whole Dot Ecosystem writes to — plus the audit trail every governance gate depends on. It's infrastructure, not a knowledge consumer.
What it does
Tiers
Actively referenced graph nodes live in Hot (≤50ms p95). Unreferenced knowledge ages into Warm, then Cold — promotion is demand-driven, demotion is policy-driven and logged.
Retrieval
Four consumer-visible retrieval classes — agent-context, surface, audit, archive — each with an explicit, specified on-breach behavior, not an afterthought.
Indexes
A graph store plus versioned vector indexes for semantic retrieval, backing every platform's Knowledge Pack archive.
Forgetting
Knowledge loses salience through dormancy and expiry, not deletion. Superseded versions demote toward Cold with the provenance chain intact.
Durability
Scheduled integrity checks and durability outcomes back every tier. Nothing skips Cold straight to deletion.
Trust
We publish latency, durability, and integrity data about our own operation — never the knowledge itself. The data plane and telemetry plane are architecturally separate.
Storage tiers
Every tier has a named latency target and a real backing store. Promotion is demand-driven; demotion is policy-driven and logged — nothing skips Cold to deletion.
| Tier | Contents | Latency | Backing |
|---|---|---|---|
| Hot | Referenced graph nodes/edges, verified lessons | ≤ 50ms p95 | Graph store + cache |
| Warm | Unreferenced 90 days–2yrs, provisional conclusions | ≤ 2s p95 | Warm store |
| Cold | Superseded/retracted knowledge, full ledger history | ≤ 5min async | Immutable archive |
Knowledge-graph storage, vector indexes, and audit trails — the substrate the rest of the Dot Ecosystem already writes to.