End-to-end latency model
Per-surface latency
| Surface | Target p50 | Target p99 | Notes |
|---|---|---|---|
| Kafka Streams | 800 ms | 2 s | Closest to source; used for low-latency pipelines |
| WebSocket | 1.0 s | 3 s | Same source as Kafka; adds fan-out overhead |
| REST (cached) | 1.5 s | 5 s | Cache TTLs vary per endpoint — see each one |
| REST (uncached) | 3 s | 10 s | Rare; for deep historical lookups |
| GraphQL | 2 s | 6 s | Joins multiple products; adds planning time |
Per-product refresh
| Product | Real-time? | Snapshot cadence |
|---|---|---|
| Tokens — market | yes | price tick every trade |
| Tokens — security | delayed | minutes after creation |
| Trades | yes | per block |
| Pools | yes | per event |
| Wallets — holdings | yes | seconds after transfer |
| Wallets — net worth history | no | hourly |
| Candles (OHLC) | yes | bar closes at interval |
| Holders (table) | yes | per transfer |
| Holders (distribution metrics) | no | every 5 min |
| Smart Money (cohort membership) | no | weekly (Mon 00:00 UTC) |
| Smart Money (flow events) | yes | per trade |
| Blockchain Core | yes | per block |
| Rankings | yes (most) | 1 min / block-level |
| Compliance (KYT / KYA) | yes | on request |
Backfill behaviour
- Solana / EVM archival data backfilled from genesis where feasible.
- New chains start with ≥ 30 days of backfill at launch, full history streamed in within the first 90 days.
- Reorgs cause rewrites of the affected slot/block in all downstream streams; consumers should be idempotent on
(chainId, txHash)+logIndex.
Incidents & status
Current incidents and historical uptime live at status.chainstream.io (link to be finalized).Next
Rate limits
Per-plan throughput limits and headers.
Kafka topic catalog
Exact topic names and retention per chain.