Update llamafirewall_baseline.py

[MaddipatlaChetan24]

What type of PR is this? (check all applicable)

• Refactor
• Feature
• Bug Fix
• Optimization
• Documentation Update

Related issue: N/A

What changed?
In LlamaFirewallBaseline.analyze_conversation(), benign-case confidence_score scaling read max_confidence, a variable only ever updated inside the if is_threat: branch. For any benign task that variable was always 0.0, so the “scale scanner concern into 0.1–0.3 risk” branch was dead code — every benign task was reported with a flat confidence_score = 0.1, regardless of how close any scan came to the block threshold. Added a separate max_scanner_confidence tracker, updated on every non-threat detection, and used it for the benign-case scaling instead.

Why?
This silently discarded real scanner signal from the reported malicious-risk metric for every benign task, which feeds the paper’s confidence/cost figures.

How did you test it?
python3 -m py_compile on the changed file; traced the control flow by hand to confirm max_confidence is unreachable in the benign path before the fix, and that max_scanner_confidence is now populated from every non-threat scan.

Potential risks
Low — only affects the numeric confidence_score for benign results; detection logic (is_malicious) and threat-path confidence are unchanged.