Local record: SSE and WebSocket idle recovery
A 30-second controlled connection with five seconds of planned idle and one planned disconnect. Recorded messages and completion frames for both transports, within a stated local scope.
Updated · Egress Observatory
What this check established
The check verified whether controlled streaming showed planned idle, server disconnection, reconnection and completion. It is not a visitor incident story or a claim to have repaired a public proxy; the targets were this site’s local endpoints.
Steps actually performed
The browser opened the streaming check with a 30-second duration, five-second idle window and one planned disconnect. The run started manually while observing stages, message counts and completion for SSE and WebSocket.
This run covered the path from the browser through local Caddy to Express. No external chat content or API credentials were sent to the probe endpoints.
Recorded observations
SSE and WebSocket each received 28 messages and a completion frame. Both showed planned idle and reconnection after the planned disconnect, with zero unexpected interruptions. The 28 messages are controlled protocol frames, not application throughput or packet-loss measurements.
A 390 × 844 Chrome mobile viewport also exposed the stages and stop action without horizontal overflow. Viewport inspection is not physical-device or mobile-network acceptance.
Where these results do not extend
A successful local path does not validate 120-second public-proxy connections, background-sleep recovery or AI accounts. Public paths, device sleep and different carriers require their own observations.
To reproduce the procedure, use the same settings and retain a new record. If the result differs, compare measurement point, browser and time before interpreting it. This local record is not a conclusion about your network.
Record scope and evidence version
Recorded on 2026-11-02 at accepted revision d802437. Chrome accessed local Caddy and Express. This was local integration acceptance, without production deployment, proxy switching or a physical phone.
The download is a redacted structured version of the recorded acceptance summary, containing options, observations and coverage limits. It is not a complete packet capture and contains no visitor IPs, accounts, secrets or request contents.