> ## Documentation Index
> Fetch the complete documentation index at: https://docs.lasso.sh/llms.txt
> Use this file to discover all available pages before exploring further.

# WebSocket continuity evidence

> Scope and result of Lasso's controlled newHeads provider-replacement soak.

Lasso tested `newHeads` subscription recovery during provider replacement on
a surviving downstream WebSocket connection. This is a controlled continuity
test, not a guarantee across client disconnections, application-node loss, or
exhausted recovery bounds. Applications should keep checkpoints and reconcile
after reconnecting. See [subscription recovery limits](/advanced/websocket-subscriptions#application-checkpoints-and-reorgs).

## Test and result

On August 22, 2026, a pinned Lasso implementation ran a 30-minute in-process
soak. Fifty downstream clients shared one upstream `newHeads` subscription.
The harness forced a provider replacement every 15 seconds and skipped three
live blocks per replacement so HTTP replay had to restore the gap. It recorded
119 replacements, zero downstream gaps, and zero replay-created duplicates.
The coordinator count stayed at one. Subscription state drained after the last
client left, within the test's 95-second teardown limit.

```text theme={null}
WS_SOAK_EVIDENCE {"duration_ms":1800000,"failovers":119,"clients":50,"gaps":0,"duplicates":0,"teardown_ms":89008,"final_head":9387,"warm_processes":660,"final_processes":660,"warm_memory_bytes":94445512,"final_memory_bytes":92216898,"peak_coordinators":1,"peak_backfill_owners":0,"peak_event_buffer":0,"peak_dedupe":155}
```

This is a Lasso-run gate against implementation commit
`89ad1a09ad88a1cf86a1dbc8ac5de6feb526b5f3`. Its Cloud harness is not
part of the public RPC Core release, so this result is reported evidence,
not a command you can reproduce from the public repository. It measures
bounded replacement within one profile and one running process. It does not
measure a general hosted capacity, SLA, or exactly-once delivery guarantee.