API degradation over 3 consecutive days caused significant financial loss

What happened

Over April 6-8, 2026, the Claude API experienced degraded performance across all three days
(incidents: xmqkb966qt1h, 124yr07585k9, lhws0phdvzz3).

During this period I was running multi-agent review workflows via Claude Code
that spawned ~48 sub-agents across Opus, Sonnet, and Haiku. The degraded API
produced inconsistent and unreliable outputs, requiring multiple re-runs across
5+ sessions over ~15 hours.

The resulting cost was substantial — both in API usage and wasted time — with
no indication in Claude Code that the API was degraded or that outputs might
be unreliable.

Request

1. Billing adjustment for API usage during the documented degradation windows
2. Consider surfacing API health status in Claude Code so users know when

output quality may be compromised before launching expensive workflows

Billing reference

Anthropic account invoices from May 2025 – present total ~€1,150.