LamChingFung-2425-Anti-FraudX

Feature: Scammer Strategy Management

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

Overview

Tracks which scam tactics the ScammerAgent has deployed across rounds and detects repetition fatigue — the natural drop in effectiveness when the same tactic is used repeatedly. Also suggests the next optimal tactic to keep the simulation realistic and varied.

Implementation

File: backend/utils/scammer_strategy_manager.py

Core Responsibilities

Function Description
Track tactic usage Records every tactic used per session with round number
Calculate fatigue multiplier Returns effectiveness reduction based on repetition count
Suggest next strategy Returns least-recently-used or unused tactic
Feed trust pipeline Fatigue multiplier applied before trust delta is calculated

Fatigue Multiplier

def get_fatigue_multiplier(self, tactic: str) -> float:
    count = self.tactic_counts.get(tactic, 0)
    if count == 0:   return 1.00   # First use — full effect
    if count == 1:   return 0.90   # Second use — 10% reduction
    if count == 2:   return 0.70   # Third use  — 30% reduction
    return 0.50                    # 4th+ use    — 50% reduction

Next Strategy Selection

def get_next_strategy(self) -> str:
    # Returns tactic with lowest usage count
    # Prioritises completely unused tactics
    # Falls back to least-used if all have been used

Tactic Categories

Defined in backend/config.py → ScamTacticsConfig:

Tactic ID Category Real-world Example
authority_claim Authority “我係警察” / “我係銀行職員”
urgency Urgency “你帳戶即將被凍結” / “限時優惠”
social_proof Social proof 假冒朋友推薦 / 假見證
reciprocity Reciprocity 免費試用 / 小禮物
scarcity Scarcity “只剩最後3個名額”
fear Fear 假冒法院通知 / 逮捕令威脅
liking Rapport 假裝共同興趣 / 情緣詐騙
commitment Commitment 逐步升級承諾
information_gathering Reconnaissance 問問題套取個人資料
financial_hook Financial 保證高回報投資

Integration with Trust Pipeline

ScammerAgent generates response
  │
  ▼
Tactic identified from response (keyword match or LLM classification)
  │
  ▼
ScammerStrategyManager.record_tactic(tactic)
ScammerStrategyManager.get_fatigue_multiplier(tactic)
  │
  ▼
Raw trust delta × fatigue_multiplier × emotional_multiplier
  │
  ▼
VictimTrustState.update(change_type, adjusted_delta, reason)

Per-Session Isolation

Each GameSession has its own ScammerStrategyManager instance. Tactic counts reset to zero at session start — fatigue does not carry over between different players or sessions.

Relevant Files

backend/utils/scammer_strategy_manager.py   Core tactic tracking + fatigue logic
backend/utils/performance_tracker.py        Applies fatigue multiplier in trust update
backend/config.py                           ScamTacticsConfig (tactic list)
backend/agents/scammer.py                   ScammerAgent that generates tactics