Skip to content

Failure diagnostics ​

Open a release's Details in the dashboard. Failure diagnostics show counts and the most recent report for each reason from the last 30 days.

ReasonMeaning
download_failedDownload could not complete, including network failure.
integrity_rejectedPackage content, hash or archive checks failed.
signature_rejectedSigning key or signature validation failed.
runtime_mismatchPackage does not match the installed native runtime/version.
startup_unacknowledgedThe previous startup did not confirm readiness before the next launch.
watchdog_timeoutThe foreground readiness watchdog expired after ten seconds.
install_failedAn installation failure reported through the diagnostic API.

Native startup and watchdog reasons are saved with failed-package metadata and retried through the existing status-report flow after recovery. Download diagnostics are best effort, with a short non-blocking reporting window; a disconnected device may never deliver that report. Missing reports do not prove a release is healthy. Telemetry/data-transmission opt-outs are respected.

Only bounded reason codes are accepted. No exception messages, stack traces or raw installation identifiers are stored. Reports are deduplicated per deployment, release, reason and hashed installation identifier. Storage keeps at most 10,000 diagnostics per deployment and 100,000 globally, and removes records older than 30 days on ingestion; the dashboard always hides expired records.

These are client-reported signals, not attested device evidence. Possession of a deployment key is enough to submit them. They do not trigger automatic fleet rollback. Use staged rollouts, release controls and independent application monitoring to decide whether to pause or roll back.

A version that fails still leaves the last healthy app usable. Publishing a mandatory update does not lock the user out when download or startup fails.

Recent delivery observations ​

The dashboard Downloads, Healthy starts and Device rollbacks totals reflect retained client reports from the last 30 days, not lifetime users. Storage is bounded to 10,000 recent observations per deployment and 100,000 globally, so busy deployments can show a shorter retained window. Reports without an installation ID are counted once per release/event kind; replaying the same anonymous report does not add users. These untrusted client signals aid investigation and do not independently prove device health.

Rollback releases is a separate server-side count of successful rollback publications for the selected deployment over the last 30 days. It includes actions through the dashboard and terminal. It is not a client health metric. See release controls and activity history.

Signed React Native updates. Self-hosted on your infrastructure.