1 Your estate
Counts × per-unit EPS (editable — defaults from the sizing guide's device catalog).
SourceCountEPS / unit
2 Posture & retention
The raw toggle is the biggest single lever — it roughly halves or doubles volume.
Your load Live rate card
Average EPS
—
—
Events / day
—
Ingested (license)
—
wire GB/day — the plan dimension
Stored in the lake
—
GB/day on disk (v2 segments)
On SaaS (we host it)
—
—
| You deploy | — |
| Egress to the ingest edge | — |
| Your compute / storage / platform ops | none — in the plan |
Region-pinned pooled cell by default; dedicated single-tenant cells (own VPC, PrivateLink) for regulated estates — see the deployment-models doc below.
Self-managed (you host it)
| Item | Sized |
|---|---|
| Brain nodes (cells) — | — |
| Per node | — |
| Cloud example / on-prem VM | — |
| Store NVMe (on-node retention) | — |
| Compliance archive (compressed) | — |
| Failover replay (RTO) per node | — |
—
Assumptions & constants (v1 · 2026-07-03 — planning defaults, measured where marked)
| Event size (wire, raw ON / OFF) | 1,475 / 903 B measured |
| Event size (stored v2, raw ON / OFF) | 1,047 / 538 B measured |
| Per-family sizes in inventory mode | measured per source family |
| Compute | 1,500 EPS / vCPU (+2 floor, min 4) |
| RAM | 8 GB + hot-window (3.5 KB/event) + 2 GB/10k EPS |
| Disk overhead / archive compression | ×1.44 / ÷3 (gzip measured 4.3×) |
| Per-brain planning envelope | 10,000 avg EPS → shard into cells |
| Platform's own events | +3 % EPS |
| Boot-replay rate (RTO math) | 50,000 events/s measured |
Constants are measured on the real event + sealed segment-store path and mirror the platform's own sizing & capacity guide §12; the published reference set is listed on Resources. The formal per-node capacity target is an open engineering ratification — these defaults deliberately derate measured ceilings. Not legal advice; your compliance team owns the retention period.