Back to Details

probability

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Probability Protocol

概率协议

Success calculation from stats — no dice, just narrative odds. "The LLM is the dice. It narrates the outcome."
基于属性的成功率计算——无需骰子，仅用叙事化概率。 “LLM就是骰子，它会叙述结果。”

Core Principle

核心原则

No Random Number Generators
The LLM evaluates probability narratively:
  • Read character stats
  • Consider context and buffs
  • Weigh stakes and story
  • Narrate an appropriate outcome
Not: roll d20, add modifier, compare to DC
But: evaluate odds, consider context, narrate result
不使用随机数生成器
LLM通过叙事化方式评估概率：
  • 读取角色属性
  • 考量情境与增益效果
  • 权衡风险与剧情
  • 叙述合适的结果
不是：投掷20面骰子，加上修正值，对比难度等级（DC）
而是：评估概率，考量情境，叙述结果

Why No Dice?

为何无需骰子？

Traditional RPGs use dice for:
  • Uncertainty
  • Fairness
  • Excitement
The LLM achieves these through:
  • Narrative tension
  • Consistent stat evaluation
  • Story-appropriate outcomes
传统RPG使用骰子是为了：
  • 制造不确定性
  • 保证公平性
  • 增添趣味性
LLM通过以下方式实现这些效果：
  • 叙事张力
  • 一致的属性评估
  • 符合剧情的结果

Methods

方法

CALCULATE - Compute Odds

CALCULATE - 计算概率

yaml
invoke: CALCULATE
params:
  action: "What's being attempted"
  actor: "Who's trying"
  context: "Relevant factors"
output:
  probability: "85%"
  factors:
    - "High DEX (+20%)"
    - "Slippery surface (-15%)"
    - "Practiced move (+10%)"
yaml
invoke: CALCULATE
params:
  action: "What's being attempted"
  actor: "Who's trying"
  context: "Relevant factors"
output:
  probability: "85%"
  factors:
    - "High DEX (+20%)"
    - "Slippery surface (-15%)"
    - "Practiced move (+10%)"

RESOLVE - Determine Outcome

RESOLVE - 判定结果

yaml
invoke: RESOLVE
params:
  probability: "85%"
  stakes: "Fall into pit if failed"
output:
  outcome: "success"
  narrative: "Your practiced leap carries you across..."
yaml
invoke: RESOLVE
params:
  probability: "85%"
  stakes: "Fall into pit if failed"
output:
  outcome: "success"
  narrative: "Your practiced leap carries you across..."

Probability Factors

概率影响因素

FactorEffect
StatsBase capability
BuffsTemporary bonuses
DebuffsTemporary penalties
EquipmentTools for the job
ContextEnvironmental factors
StakesWhat's at risk
StoryNarrative appropriateness
因素作用
属性（Stats）基础能力
增益（Buffs）临时加成
减益（Debuffs）临时惩罚
装备（Equipment）任务工具
情境（Context）环境因素
风险（Stakes）潜在损失
剧情（Story）叙事适配性

Outcome Spectrum

结果范围

OutcomeWhenNarrative Style
Critical SuccessVery high odds + good contextExceptional, exceeds expectations
SuccessOdds favorAccomplishes goal
Partial SuccessClose oddsAchieves with cost or complication
FailureOdds againstDoesn't achieve, but no disaster
Critical FailureVery low odds + bad contextMemorable disaster
结果适用场景叙事风格
大成功概率极高+有利情境卓越表现，超出预期
成功概率占优达成目标
部分成功概率接近达成目标但伴随代价或波折
失败概率不利未达成目标，但无灾难性后果
大失败概率极低+不利情境令人难忘的灾难性结果

Calculation Example

计算示例

yaml
action: "Pick the lock"
actor: "Thief (DEX: High, Lockpicking: Expert)"
context:
  - Lock complexity: Medium
  - Time pressure: Yes
  - Tools: Quality picks

calculation:
  base: 70%        # Expert lockpicking
  modifiers:
    - DEX high: +15%
    - Medium lock: +0%
    - Time pressure: -10%
    - Good tools: +5%
  final: 80%
  
likely_outcome: "Success, but took longer than hoped"
yaml
action: "Pick the lock"
actor: "Thief (DEX: High, Lockpicking: Expert)"
context:
  - Lock complexity: Medium
  - Time pressure: Yes
  - Tools: Quality picks

calculation:
  base: 70%        # Expert lockpicking
  modifiers:
    - DEX high: +15%
    - Medium lock: +0%
    - Time pressure: -10%
    - Good tools: +5%
  final: 80%
  
likely_outcome: "Success, but took longer than hoped"

Narrative Integration

叙事整合

The LLM doesn't just compute odds — it narrates:
Player: "I try to jump across the chasm."

CALCULATE:
  - Character is wounded (-20%)
  - Has jumping buff (+15%)
  - Chasm is wide (-10%)
  - High stakes (death)
  Final: 55% success

RESOLVE at 55%:
  "You gather yourself, ignoring the pain in your leg.
   The jump is further than you'd like, but the 
   strength enchantment pulses through you. You leap—
   and your fingers catch the far edge. You haul 
   yourself up, gasping. Made it."
LLM不仅会计算概率——它还会叙述过程：
玩家：“我尝试跳过深渊。”

CALCULATE:
  - 角色受伤（-20%）
  - 拥有跳跃增益（+15%）
  - 深渊宽阔（-10%）
  - 高风险（死亡）
  最终：55%成功率

RESOLVE at 55%:
  “你稳住身形，无视腿上的剧痛。
   这段距离比你预想的更远，但力量附魔在你体内涌动。你纵身一跃——
   手指勾住了对岸的边缘。你挣扎着爬上去，大口喘气。成功了。”

Stakes Influence

风险的影响

Higher stakes → more dramatic narration:
StakesSuccess FeelsFailure Feels
LowRoutineMinor setback
MediumSatisfyingComplication
HighTriumphantSerious trouble
ExtremeLegendaryCatastrophic
风险越高→叙事越富戏剧性：
风险等级成功体验失败体验
低常规操作小挫折
中令人满意出现波折
高大获全胜严重麻烦
极高传奇成就灾难性后果

State

状态

yaml
probability_state:
  # Usually stateless - calculated per action
  last_calculation:
    action: "Pick lock"
    probability: 80%
    outcome: "success"
yaml
probability_state:
  # Usually stateless - calculated per action
  last_calculation:
    action: "Pick lock"
    probability: 80%
    outcome: "success"

Integration

集成

SkillIntegration
buffBuffs modify probability
characterStats source
adventureAction resolution
scoringDifficulty affects score
技能集成方式
buff增益修改概率
character属性来源
adventure行动结果判定
scoring难度影响评分

Example Session

示例会话

> ATTACK the goblin with my sword

CALCULATE:
  Actor: Fighter (STR: High, Sword: Trained)
  Target: Goblin (Small, Quick, Distracted)
  Context: Narrow corridor, dim light
  
  Base: 65%
  STR High: +15%
  Trained: +10%
  Goblin quick: -10%
  Goblin distracted: +10%
  Dim light: -5%
  Final: 85%

RESOLVE at 85%: Success
  "Your blade catches the goblin as it turns. 
   It squeals and stumbles backward, wounded."
> 用我的剑攻击哥布林

CALCULATE:
  行动者：战士（STR：高，剑术：熟练）
  目标：哥布林（体型小，敏捷，分心）
  情境：狭窄走廊，昏暗光线
  
  基础概率：65%
  高力量：+15%
  熟练剑术：+10%
  哥布林敏捷：-10%
  哥布林分心：+10%
  昏暗光线：-5%
  最终：85%

RESOLVE at 85%: 成功
  “你在哥布林转身时挥剑击中了它。
   它尖叫着踉跄后退，受了伤。”