My session showed 0% context and 0 compactions on startup, even though I didn't called compact. The previous session was active and had significant context. Gateway logs from around the same time:
11:17:46 [agent/embedded] Profile anthropic:default timed out. Trying next account...
11:17:46 [agent/embedded] embedded run failover decision: runId=4d853356-c495-46ef-8bcc-1812bf2a0c8c stage=assistant decision=fallback_model reason=timeout provider=anthropic/claude-haiku-4-5 profile=sha256:154a23a3efe6
11:17:46 [diagnostic] lane task error: lane=nested durationMs=22656 error="FailoverError: LLM request timed out."
11:17:46 [diagnostic] lane task error: lane=session:agent:main:cron:c3cc85ab-54b9-41c6-b4f8-f2000bcd7006 durationMs=22660 error="FailoverError: LLM request timed out."
11:17:46 [model-fallback/decision] model fallback decision: decision=candidate_failed requested=anthropic/claude-haiku-4-5 candidate=anthropic/claude-haiku-4-5 reason=unknown next=openai/gpt-5.3-codex
11:17:46 [model-fallback/decision] model fallback decision: decision=candidate_failed requested=anthropic/claude-haiku-4-5 candidate=openai/gpt-5.3-codex reason=timeout next=anthropic/claude-sonnet-4-6
11:17:46 [model-fallback/decision] model fallback decision: decision=candidate_failed requested=anthropic/claude-haiku-4-5 candidate=anthropic/claude-sonnet-4-6 reason=timeout next=none
Question: Can an LLM request timeout cascade cause the session context to be reset/lost? Or is there another mechanism that could explain this?