LamChingFung-2425-Anti-FraudX

Feature: Trust Meter System

Category: Core Version: 4.1 Last Updated: 2026-03-11

Overview

The Trust Meter tracks the victim’s psychological state across three real-time dimensions. These values drive the outcome of every battle session — a scammer win, expert win, or victim self-protection. The system models realistic human behaviour: trust that has built up is hard to destroy, repeated tactics wear thin, and emotional states amplify or dampen persuasion.

Trust Dimensions

Dimension Range Meaning Initial Value (varies by persona)
scammer_trust 0–100 How much the victim believes the scammer 30–70
expert_trust 0–100 How much the victim trusts the expert’s advice 40–60
alertness 0–100 Victim’s overall fraud suspicion level 30–70

State Object: VictimTrustState

File: backend/utils/performance_tracker.py

@dataclass
class VictimTrustState:
    trust_in_scammer: int = 50
    trust_in_expert: int = 50
    alertness: int = 50
    emotional_state: str = "neutral"  # neutral | calm | anxious | panicked | suspicious
    history: List[Dict] = field(default_factory=list)

    def update(self, change_type: str, change_value: int, reason: str):
        # Clamps to 0–100, appends to history
        ...

The history list records every trust change with timestamp, type, value, and reason — used by RecorderAgent for post-game analysis.

Session Object: GameSession

File: backend/utils/rpgv2_game_mode_manager.py

@dataclass
class GameSession:
    session_id: str
    mode: GameMode           # "victim" | "expert" | "scammer" | "auto"
    scam_type: str
    victim_persona: str
    conversation_history: List[Dict]
    round_count: int
    trust_in_scammer: float
    trust_in_expert: float
    alertness: float
    player_score: int
    ai_score: int
    start_time: str
    last_update: str
    game_over: bool = False
    winner: Optional[str] = None
    round_logs: List[Dict] = field(default_factory=list)

round_logs captures per-round detail (agent responses, trust deltas, emotional state) for the Recorder.

Win/Loss Outcome Thresholds

Defined in backend/config.py → TrustConfig:

Condition Outcome Winner
scammer_trust >= 80 Victim scammed "scammer"
expert_trust >= 75 Victim protected "expert"
scammer_trust < 40 AND expert_trust > 60 Expert wins early "expert"
alertness >= 80 Victim self-protects "victim"
round_count >= 15 Max rounds reached No winner (draw)

Checked after every round in rpgv2_game_modes_routes.py.

Trust Modifiers

1. Inertia

High existing trust is harder to move:

Existing trust Change multiplier
≥ 80 −40% to −50%
50–79 No adjustment
≤ 20 Slightly easier to change

2. Tactic Fatigue

Repeated use of the same scam tactic loses effectiveness:

Repetition count Effectiveness multiplier
1st use 100%
2nd use 90%
3rd use 70%
4th+ use 50%

Managed by ScammerStrategyManager (backend/utils/scammer_strategy_manager.py).

3. Emotional State Multipliers

Emotional State Scammer effectiveness Expert effectiveness
anxious +30% −20%
panicked +50% −40%
suspicious −30% +40%
calm Baseline Baseline
neutral Baseline Baseline

4. Persona-Based Weight Adjustment

AdaptiveWeightOptimizer (backend/utils/adaptive_scoring.py) adjusts the weight of each evaluation dimension (empathy, clarity, evidence, actionability) based on victim persona. See adaptive-scoring.md.

Frontend Components

File Purpose
rpg-platform-v2/src/systems/TrustSystem.ts Applies trust delta received from API to local state
rpg-platform-v2/src/ui/TrustMeter.ts Renders animated progress bars in BattleScene sidebar

The frontend trust state is not authoritative — it mirrors the backend GameSession values. On restore from sessionStorage, trust values are re-synced from saved state.

Relevant Files

backend/utils/performance_tracker.py      VictimTrustState dataclass + update logic
backend/utils/rpgv2_game_mode_manager.py  GameSession dataclass + session store
backend/utils/adaptive_scoring.py         Persona-based weight adjustment
backend/utils/scammer_strategy_manager.py Tactic fatigue tracking
backend/config.py                         TrustConfig (thresholds, multipliers)
backend/api/rpgv2_game_modes_routes.py    Outcome checking per round
rpg-platform-v2/src/systems/TrustSystem.ts
rpg-platform-v2/src/ui/TrustMeter.ts