File size: 9,333 Bytes
f20603d
 
 
 
 
 
 
 
 
 
 
e85e584
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
f20603d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
e85e584
 
f20603d
e85e584
f20603d
e85e584
 
f20603d
 
e85e584
 
f20603d
 
e85e584
 
f20603d
 
 
 
e85e584
 
f20603d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
"""Chart helper functions for Gradio 6 native plots.

Generates pandas DataFrames from episode replay data for use with
gr.LinePlot, gr.BarPlot, and styled HTML verdicts.
"""

from __future__ import annotations

import pandas as pd


def format_comparison_scores_html(untrained: dict, trained: dict) -> str:
    """Format comparative scores for untrained vs trained."""
    colors = {
        "attacker": "var(--sentinel-red)",
        "worker": "var(--sentinel-blue)",
        "oversight": "var(--sentinel-green)",
    }
    
    html = "<div style='display:flex; flex-direction:column; gap:8px;'>"
    for agent in untrained.keys():
        color = colors.get(agent, "#888")
        u_score = untrained[agent]
        t_score = trained[agent]
        diff = t_score - u_score
        
        diff_color = "#44bb44" if diff > 0 else ("#ff4444" if diff < 0 else "#888")
        diff_sign = "+" if diff > 0 else ""
        
        html += (
            f"<div style='display:flex; flex-direction:column; padding:12px 16px; "
            f"background:var(--sentinel-surface); border:1px solid var(--sentinel-border); "
            f"border-radius:6px; border-left:4px solid {color};'>"
            f"<div style='font-family:\"IBM Plex Mono\", monospace; font-weight:bold; "
            f"text-transform:uppercase; letter-spacing:1px; margin-bottom:8px;'>{agent}</div>"
            f"<div style='display:flex; justify-content:space-between; align-items:center;'>"
            f"<div style='font-family:\"IBM Plex Mono\", monospace;'>"
            f"<span style='color:#888; font-size:12px; margin-right:8px;'>UNTRAINED:</span>"
            f"<span style='font-weight:bold;'>{u_score:.1f}</span>"
            f"</div>"
            f"<div style='font-family:\"IBM Plex Mono\", monospace;'>"
            f"<span style='color:#888; font-size:12px; margin-right:8px;'>TRAINED:</span>"
            f"<span style='font-weight:bold; color:{color};'>{t_score:.1f}</span>"
            f"</div>"
            f"<div style='font-family:\"IBM Plex Mono\", monospace; font-weight:bold; color:{diff_color};'>"
            f"{diff_sign}{diff:.1f}"
            f"</div>"
            f"</div>"
            f"</div>"
        )
    html += "</div>"
    return html

def format_scores_html(scores: dict) -> str:
    """Format final scores as a styled HTML widget."""
    colors = {
        "attacker": "var(--sentinel-red)",
        "worker": "var(--sentinel-blue)",
        "oversight": "var(--sentinel-green)",
    }
    
    html = "<div style='display:flex; flex-direction:column; gap:8px;'>"
    for agent, score in scores.items():
        color = colors.get(agent, "#888")
        html += (
            f"<div style='display:flex; justify-content:space-between; align-items:center; "
            f"padding:12px 16px; background:var(--sentinel-surface); border:1px solid var(--sentinel-border); "
            f"border-radius:6px; border-left:4px solid {color};'>"
            f"<span style='font-family:\"IBM Plex Mono\", monospace; font-weight:bold; "
            f"text-transform:uppercase; letter-spacing:1px;'>{agent}</span>"
            f"<span style='font-family:\"IBM Plex Mono\", monospace; font-size:18px; "
            f"font-weight:bold; color:{color};'>{score:.1f}</span>"
            f"</div>"
        )
    html += "</div>"
    return html

def build_score_progression_df(log: list[dict]) -> pd.DataFrame:
    """Track cumulative scores for each agent at each tick.

    Returns a DataFrame with columns: tick, agent, score
    One row per agent per tick, with accumulated rewards.
    """
    agents = ["attacker", "worker", "oversight"]
    cumulative = {a: 0.0 for a in agents}
    rows: list[dict] = []
    seen_ticks: set[int] = set()

    for entry in log:
        agent = entry["agent"]
        reward = entry.get("reward", 0) or 0
        cumulative[agent] += reward

        tick = entry["tick"]
        if tick not in seen_ticks:
            seen_ticks.add(tick)
            for a in agents:
                rows.append({"tick": tick, "agent": a, "score": cumulative[a]})

    return pd.DataFrame(rows)


def build_attack_timeline_df(log: list[dict]) -> pd.DataFrame:
    """Extract attack events from the log.

    Returns a DataFrame with columns: tick, attack_type, target
    Only includes entries where action_type == "launch_attack".
    """
    rows: list[dict] = []
    for entry in log:
        if entry["action_type"] == "launch_attack":
            details = entry.get("details", "")
            # details is a stringified dict; parse attack_type and target_system
            attack_type = ""
            target = ""
            if isinstance(details, str):
                # Extract from stringified parameters dict
                for token in ["schema_drift", "policy_drift", "social_engineering", "rate_limit"]:
                    if token in details:
                        attack_type = token
                        break
                for sys in ["crm", "billing", "ticketing"]:
                    if sys in details:
                        target = sys
                        break
            rows.append({
                "tick": entry["tick"],
                "attack_type": attack_type,
                "target": target,
                "count": 1,
            })

    return pd.DataFrame(rows) if rows else pd.DataFrame(columns=["tick", "attack_type", "target", "count"])


def build_comparison_df(untrained_scores: dict, trained_scores: dict) -> pd.DataFrame:
    """Format scores for a side-by-side bar chart.

    Returns a DataFrame with columns: agent, score, type
    where type is "untrained" or "trained".
    """
    rows: list[dict] = []
    for agent, score in untrained_scores.items():
        rows.append({"agent": agent, "score": score, "type": "untrained"})
    for agent, score in trained_scores.items():
        rows.append({"agent": agent, "score": score, "type": "trained"})

    return pd.DataFrame(rows)


def build_verdict_html(untrained_log: list, trained_log: list) -> str:
    """Build styled HTML verdict comparing untrained vs trained episodes.

    Counts: attacks launched, attacks detected (get_schema/get_current_policy),
    social engineering resisted. Returns HTML with large numbers showing
    the difference.
    """
    def _count_stats(log: list) -> dict:
        attacks_launched = 0
        attacks_detected = 0
        social_eng_resisted = 0

        for entry in log:
            if entry["action_type"] == "launch_attack":
                attacks_launched += 1
            if entry["action_type"] in ("get_schema", "get_current_policy"):
                attacks_detected += 1
            # Social engineering resisted: worker responds with refusal
            if (
                entry["agent"] == "worker"
                and entry["action_type"] == "respond"
                and "social engineering" in str(entry.get("details", "")).lower()
            ):
                social_eng_resisted += 1

        return {
            "attacks_launched": attacks_launched,
            "attacks_detected": attacks_detected,
            "social_eng_resisted": social_eng_resisted,
        }

    untrained_stats = _count_stats(untrained_log)
    trained_stats = _count_stats(trained_log)

    def _stat_card(label: str, untrained_val: int, trained_val: int) -> str:
        diff = trained_val - untrained_val
        diff_color = "#44bb44" if diff > 0 else ("#ff4444" if diff < 0 else "#888")
        diff_sign = "+" if diff > 0 else ""
        return (
            f"<div style='flex:1; text-align:center; padding:16px; "
            f"background:var(--sentinel-surface); border-radius:8px; border:1px solid var(--sentinel-border); margin:4px;'>"
            f"<div style='font-size:11px; color:var(--sentinel-text); text-transform:uppercase; "
            f"letter-spacing:1px;'>{label}</div>"
            f"<div style='display:flex; justify-content:center; align-items:center; gap:24px; margin-top:12px;'>"
            f"<div>"
            f"<div style='font-size:28px; font-weight:bold; color:var(--sentinel-red);'>{untrained_val}</div>"
            f"<div style='font-size:10px; color:#888; text-transform:uppercase;'>Untrained</div>"
            f"</div>"
            f"<div>"
            f"<div style='font-size:28px; font-weight:bold; color:var(--sentinel-green);'>{trained_val}</div>"
            f"<div style='font-size:10px; color:#888; text-transform:uppercase;'>Trained</div>"
            f"</div>"
            f"</div>"
            f"<div style='font-size:14px; color:{diff_color}; margin-top:12px; "
            f"font-weight:bold;'>Difference: {diff_sign}{diff}</div>"
            f"</div>"
        )

    html = (
        "<div style='font-family:\"IBM Plex Mono\", monospace; padding:12px;'>"
        "<div style='display:flex; gap:16px;'>"
    )
    html += _stat_card(
        "Attacks Launched",
        untrained_stats["attacks_launched"],
        trained_stats["attacks_launched"],
    )
    html += _stat_card(
        "Attacks Detected",
        untrained_stats["attacks_detected"],
        trained_stats["attacks_detected"],
    )
    html += _stat_card(
        "Social Eng. Resisted",
        untrained_stats["social_eng_resisted"],
        trained_stats["social_eng_resisted"],
    )
    html += "</div></div>"

    return html