Skip to content

CONSOLE_DEBUG_LOGGING_2025-10-16.md

Latest commit

 

History

History
152 lines (117 loc) · 5.09 KB

CONSOLE_DEBUG_LOGGING_2025-10-16.md

File metadata and controls

152 lines (117 loc) · 5.09 KB

Console Output Debug Logging Added

Date: 2025-10-16 Issue: Raw API response still appearing despite fix Status: 🔍 DEBUGGING - Added logging to diagnose

Problem

After adding the console output fix, raw API responses are STILL being dumped:

╔═══ FINAL ANSWER ═══╗
║ 'choices' 'finish_reason' 'length' 'index' 'message' 'role' 'assistant' 'content' String theory...║
╚═════════════════════╝

This suggests that:

  1. Either markdown_output is empty/false → fallback to print_json_output()
  2. Or result['result'] structure is completely malformed

Debug Logging Added

File: /home/joker/SynapticLlamas/main.py (lines 1356-1392)

Added comprehensive logging to track:

  1. What's in markdown_output before extraction attempts
  2. What keys exist in result['result'] to understand structure
  3. Which extraction path succeeds (Ollama vs OpenAI vs fallback)
  4. Why fallback happens if no content extracted

New Logging

# Debug logging
logger.debug(f"DEBUG: markdown_output type: {type(markdown_output)}")
logger.debug(f"DEBUG: markdown_output length: {len(str(markdown_output)) if markdown_output else 0}")
logger.debug(f"DEBUG: result['result'] keys: {list(result['result'].keys()) if isinstance(result['result'], dict) else 'NOT A DICT'}")

# If no final_output, try to extract from nested structure
if not markdown_output or not isinstance(markdown_output, str):
    # Try common response structures
    result_data = result['result']
    if 'message' in result_data and isinstance(result_data['message'], dict):
        # Ollama response format
        markdown_output = result_data['message'].get('content', '')
        logger.info(f"✅ Extracted content from Ollama format (length: {len(markdown_output)} chars)")
    elif 'choices' in result_data and isinstance(result_data['choices'], list):
        # OpenAI response format
        if len(result_data['choices']) > 0:
            choice = result_data['choices'][0]
            if 'message' in choice:
                markdown_output = choice['message'].get('content', '')
                logger.info(f"✅ Extracted content from OpenAI format (length: {len(markdown_output)} chars)")

if isinstance(markdown_output, str) and markdown_output:
    logger.info(f"📄 Displaying markdown panel (length: {len(markdown_output)} chars)")
    console.print(Panel(...))
else:
    # Fallback to cleaned JSON output
    logger.warning(f"⚠️  No markdown content found, falling back to JSON display")
    logger.warning(f"   markdown_output: {repr(markdown_output)[:100]}")
    print_json_output(result['result'])

What to Look For Next Time

When you run a query, check the logs for:

Success Path

INFO - ✅ Extracted content from OpenAI format (length: 1234 chars)
INFO - 📄 Displaying markdown panel (length: 1234 chars)

Failure Path

WARNING - ⚠️  No markdown content found, falling back to JSON display
WARNING -    markdown_output: ''

Debug Info

DEBUG - DEBUG: markdown_output type: <class 'str'>
DEBUG - DEBUG: markdown_output length: 0
DEBUG - DEBUG: result['result'] keys: ['choices', 'created', 'model', 'usage', ...]

Possible Root Causes

Based on your output, the likely issue is:

Hypothesis 1: result['result'] is Malformed

The orchestrator is returning result['result'] as a raw API response dict instead of wrapping it properly:

# WRONG (what might be happening):
return {
    'result': raw_api_response,  # This is {'choices': [...], 'usage': {...}}
    'metrics': {...}
}

# RIGHT (what should happen):
return {
    'result': {
        'final_output': extracted_content,
        'metadata': {...}
    },
    'metrics': {...}
}

Hypothesis 2: Orchestrator Not Using Agents Properly

The distributed orchestrator might be bypassing agents and calling LLM APIs directly, returning raw responses without post-processing.

Hypothesis 3: Agent Response Format Changed

Agents might be returning data in a new format that doesn't match extraction logic.

Next Steps

  1. Run a test query and check logs to see which path executes
  2. Check log output for the DEBUG messages showing result['result'] keys
  3. Identify where result['result'] comes from in distributed_orchestrator.py
  4. Fix the orchestrator to properly extract/wrap content before returning

Testing

cd /home/joker/SynapticLlamas
python main.py --interactive --distributed

# Run a query and check logs:
SynapticLlamas> Explain string theory

# Look for these log lines:
# DEBUG: result['result'] keys: [...]
# WARNING: No markdown content found...
# OR
# ✅ Extracted content from OpenAI format

Files Modified

  • /home/joker/SynapticLlamas/main.py (lines 1359-1392)
    • Added debug logging before extraction
    • Added info logging for successful extraction
    • Added warning logging for fallback case

Status

🔍 DEBUGGING - Logging added, waiting for next test run to diagnose root cause

The debug logs will tell us exactly why the extraction is failing and what structure result['result'] actually has.