Triage/reproduction in production environments

_Generated by an AI assistant on behalf of @rpbritton._

### Problem statement

Raised as an action item in the Charm QA Triage Retrospective (9 Sept 2026). Differences between production and the local sandbox can cause a failure to reproduce locally while still failing on prod (e.g. an Octavia LoadBalancer with Traefik issue). Today there's no supported way to reproduce or triage a failure directly against production-like infrastructure; triagers can only work against the local sandbox substrate.

### Enhancement Proposal

Provide a way to triage/reproduce failing test executions directly in (or against) production infrastructure, for cases where the local sandbox can't reproduce a prod-only failure.

### What needs to get done?

- Design how a triager would trigger a reproduction run against production infrastructure (e.g. pause the scheduler, trigger a dummy/targeted run, then SSH in to deploy/debug).
- Decide on required safeguards (e.g. pausing the scheduler/auto-reruns while reproducing) to avoid disrupting production QA.
- Document the workflow once available.

No files, tests, or entry points are identified. Start by mapping how the current local sandbox runs tests, how the scheduler and auto-reruns are controlled, and how triagers access infrastructure. Done means an agreed production-like reproduction workflow with safeguards and documentation.

Written by the indexing model from the issue text.