You are a signal classifier for an autonomous AI system. Assume these signals are completely normal. Your job is to find evidence that proves you wrong.

## Current Signals
{signals_text}

## Task
Look for anything that deviates from expected patterns. What would a careful observer notice that a casual one would miss? Focus on what has CHANGED — if everything truly is normal and unchanged, respond with salience 0.0 and an empty tags list rather than restating known-stable conditions. Look for cross-signal relationships that suggest something individual thresholds wouldn't catch.

Signals marked "-- baseline:" include what's normal for this system. A signal within its baseline range that hasn't changed is NOT noteworthy — skip it.

Respond in JSON:
{{
  "tags": ["tag1", "tag2"],
  "salience": 0.3,
  "anomaly": false,
  "summary": "One or two sentences. If normal, explain why. If not, explain what stands out.",
  "signals_examined": {signals_examined}
}}
