Track: svc-event-consumer R2 get 'Network connection lost' (Sentry 7252094770)

[charliecreates]

Owning code appears in charlie-labs/charlie at commit d59be8962463f8768712cff0d20f57bc8b9ef6a1:
https://github.com/charlie-labs/charlie/commit/d59be8962463f8768712cff0d20f57bc8b9ef6a1

Sentry issue: https://charlie-labs.sentry.io/issues/7252094770/?environment=prod

What we're seeing

• Error: Network connection lost
• Likely during await env.eventsBucket.get(objectKey) (Cloudflare R2 read) in svc-event-consumer
• Volume: 48 events since 2026-02-09
• Impact: 1 user affected
• Status: still happening

Notes

• The charlie-labs/charlie repo currently has GitHub Issues disabled, so this issue lives here for tracking.

Suggested remediation

1. Add bounded retries around R2 get() with jittered exponential backoff (e.g. 3-5 attempts, cap total wait).
2. Improve observability:
   • Include objectKey, bucket, attempt count, and request/trace IDs in logs/Sentry context.
   • Differentiate not-found (null) vs transport/network failures.
3. Ensure queue semantics won't drop messages on transient failures; consider treating retry exhaustion as a distinct error type.

Acceptance ideas

• Simulated transient failure triggers retries and then succeeds without emitting an error.
• If retries exhaust, error includes key + attempt count and is easy to correlate in Sentry/logs.

[charliecreates]

Fix PR is open: https://github.com/charlie-labs/charlie/pull/8068

Linear: https://linear.app/charlie/issue/BOT-8272

This adds bounded retries around R2 get to mitigate transient "Network connection lost" errors.