Problem

When the gateway's failover system exhausts all auth profiles and model fallbacks, the resulting error log entries are missing provider and session_id metadata. This makes it difficult to attribute errors during post-mortem analysis.

Evidence

Analyzing the last 100 errors from our api_health_log table (structured ingestion from gateway.err.log):

• 18 of 100 errors (18%) had no provider or session_id populated
• These are all failover/wrapper errors — the final FailoverError after all profiles/models have been tried
• The underlying per-profile errors DO have provider attribution, but the wrapper error that gets logged does not carry it through

Expected behavior

Failover wrapper errors should include:

1. The provider(s) attempted (even if multiple)
2. The session_id / lane that originated the request
3. Consistent session_id format (currently we see both session:browser-XXXX and browser-XXXX for the same session)

Impact

When triaging rate limit storms or auth failures, the missing attribution means you have to manually correlate wrapper errors with nearby per-profile errors by timestamp. At scale (we had 6,000+ rate limit errors in 12 days), this makes automated analysis unreliable.

Environment

• OpenClaw version: 2026.3.7
• Providers: anthropic (token), openai-codex (OAuth)
• Config: 3 anthropic profiles, 1 openai-codex profile, 2 model fallbacks

Suggestion

In the FailoverError constructor or the error logging path, propagate the provider from the last-attempted profile and the sessionId / lane from the originating request. For the session_id normalization, ensure the session: prefix is consistently applied.