1 / 9
Bittensor Subnet Ideathon | Bittensor 子网创意马拉松

MiroSwarm

Decentralized Event Simulation & Prediction Engine
去中心化事件模拟与预测引擎

3 Time Windows 时间窗口
4 Scoring Dimensions 评分维度
5 Anti-Gaming Layers 防作弊层
2 Output Tracks 双轨输出
Origin | 起源

What Is Mirofish? And Why We Need MiroSwarm
Mirofish 是什么?为什么需要 MiroSwarm?

📘 Mirofish (Original) 原版

An AI-powered event simulation platform that analyzes news events and generates multi-horizon prediction reports (2-day, 7-day, 14-day).
一个 AI 驱动的事件模拟平台,分析新闻事件并生成多时间跨度预测报告。

  • Geopolitical, financial, tech, social event coverage
    覆盖地缘政治、金融、科技、社会事件
  • Cascading reasoning: event → impact chains → scenario branches
    级联推理:事件 → 影响链 → 情景分支
  • Professional report format with confidence ratings
    专业报告格式,带置信度评级

⚠️ Critical Weaknesses 关键缺陷

  • Centralized — single entity controls model, data, and methodology
    中心化 — 单一实体控制模型、数据和方法论
  • Single Method — only one forecasting approach, no competitive evolution
    单一方法 — 只有一种预测方法,没有竞争进化
  • No Validation Loop — predictions are never systematically scored against real outcomes
    无验证闭环 — 预测从未被系统性地与真实结果对比打分
  • No Incentive Engine — no mechanism to attract, reward, or filter forecasters
    无激励机制 — 没有吸引、奖励或淘汰预测者的机制
  • Cannot Scale — limited by internal team capacity, not by market demand
    无法规模化 — 受限于内部团队能力,而非市场需求
MiroSwarm = Mirofish's Core Concept + Bittensor's Decentralized Incentive Mechanism
MiroSwarm = Mirofish 的核心概念 + Bittensor 的去中心化激励机制
For Miners | 矿工视角

Miner Task: Predict & Produce Dual-Track Output
矿工任务:生成双轨预测输出

MINER 矿工

Track A — Structured Data 结构化数据

For validator scoring. Machine-readable, instantly comparable.
供验证者评分,机器可读,即时比对。

{
  "direction": "escalate",
  "milestones": [...],
  "confidence": 0.72,
  "coverage": ["economic", "political"]
}
  • Must submit SHA-256 hash on-chain at T+0
    必须在 T+0 将哈希上链
  • Reveal original text at T+14 for verification
    T+14 揭示原文供验证
MINER 矿工

Track B — Professional Report 专业报告

For paying customers. Readable, insightful, decision-ready.
供付费客户消费,可读、有洞察、决策就绪。

  • Executive summary with core judgment
    执行摘要 + 核心判断
  • Window-by-window scenario analysis
    分窗口情景分析(2日/7日/14日)
  • Causal logic & methodology notes
    因果逻辑 + 方法论说明
  • Limitations & monitoring points
    局限性声明 + 关键监测点

Critical: Both tracks must be content-consistent. Discrepancies are auto-detected and penalized.
关键:双轨内容必须一致,不一致将被自动检测并惩罚。

For Validators | 验证者视角

Validator Task: Feed, Verify & Score
验证者任务:投喂事件、采集真相、客观评分

VALIDATOR 验证者

Step 1: Event Distribution 事件分发

Draw events from candidate pool using block-hash random seed. Broadcast to all miners.
使用区块哈希随机种子从候选池抽取事件,广播给所有矿工。

VALIDATOR 验证者

Step 2: Ground-Truth Collection 真相采集

At T+2 / T+7 / T+14, collect real-world outcomes via news APIs, social trends, market data + 3-source cross-validation.
在三个时间点采集真实走向,新闻 API + 社交媒体 + 市场数据,3 源交叉验证。

VALIDATOR 验证者

Step 3: Four-Dimensional Scoring 四维评分

Run the scoring engine, execute consistency-anchor scan, compute composite score, submit weights to Yuma Consensus.
运行评分引擎,执行一致性锚定扫描,计算综合得分,提交权重。

Score = 0.50 × Accuracy + 0.25 × Coverage + 0.15 × Calibration + 0.10 × ReportQuality
Scoring | 评分机制

How Validators Score Miners | 验证者如何给矿工打分

MINER OUTPUT 矿工输出

Accuracy 50% — 预测准确度

Direction (50%) + Milestone Hit (30%) + Causal Match (20%)
方向正确性 + 关键节点命中 + 因果链匹配

2-day 20% / 7-day 35% / 14-day 45% window weighting
三窗口加权

MINER OUTPUT 矿工输出

Coverage 25% — 维度覆盖率

Economic, political, social, tech, financial dimensions covered vs. actual impact.
经济、政治、社会、技术、金融等维度覆盖 vs 实际影响。

MINER OUTPUT 矿工输出

Calibration 15% — 置信度校准

Brier Score punishes overconfidence. Forces honest uncertainty assessment.
Brier Score 惩罚过度自信,迫使诚实评估不确定性。

Example: Brier 0.18 → Score 0.78
示例:Brier 0.18 对应得分 0.78

MINER OUTPUT 矿工输出

Report Quality 10% — 报告质量

Consistency (40%) + Logic (25%) + Insight (20%) + Readability (15%)
内容一致性 + 逻辑清晰度 + 洞察深度 + 可读性

Security | 防作弊机制

5 Layers: What Miners Must Do vs What Validators Enforce
五层防御:矿工必须遵守 + 验证者强制执行

Layer Miner Obligation 矿工义务 Validator Enforcement 验证者执行
1 Commit-Reveal: Submit SHA-256 hash at T+0, reveal at T+14
T+0 上链哈希,T+14 揭示原文
Recompute hash, compare on-chain. Mismatch = all scores zeroed.
重新计算哈希比对,不匹配则该期所有得分清零。
2 Cannot choose events. Receive whatever block-hash seed provides.
无法选择事件,区块哈希随机分配。
Publish seed derivation + sampling algorithm. Auditable by anyone.
公开种子推导和采样算法,任何人可审计。
3 Must accept multiple validators' independent scores.
必须接受多个验证者的独立评分。
3+ validators score independently. Yuma Consensus trims outliers.
3+ 验证者独立评分,Yuma 修剪离群值。
4 Must declare honest confidence. Cannot default to 99%.
必须诚实声明置信度,不能默认 99%。
Brier Score > 0.40 → Calibration score < 0.39. 3x consecutive → reputation decay.
Brier > 0.40 严重拉低 15% 权重,连续 3 次触发声誉衰减。
5 Structured JSON and report must tell the same story.
结构化数据和报告必须讲同一个故事。
NLP scan: embedding_similarity × keyword_match_rate < 90% = 0 on consistency.
NLP 扫描语义匹配度 < 90% 则一致性子维度得 0 分。
Incentives | 激励机制

How Miners Earn vs How Validators Earn
矿工如何赚钱 vs 验证者如何赚钱

MINER 矿工

Emission Rewards 排放奖励

  • Breakthrough: New benchmark = up to 80% of epoch emissions
    突破基准可获得当期 80% 排放
  • Sustained: Near-benchmark miners share remaining pool proportionally
    接近基准按比例分配剩余排放
  • Organic: 90–95% of API/subscription revenue flows to miner whose predictions were purchased
    90-95% API/订阅收入流向被消费的预测矿工
VALIDATOR 验证者

Bond & Consensus Rewards 债券与共识奖励

  • Bonds appreciate when high-quality miners are discovered early
    早期发现优质矿工,债券升值
  • Dividends scale with consensus alignment — outliers get zeroed
    与共识对齐度决定分红,离群者归零
  • Subnet treasury (5–10% of organic revenue) funds validator ops + development
    子网国库(有机收入 5-10%)资助验证者运营与开发
Standard Bittensor Split: ~41% Miners | ~41% Validators (via Bonds) | ~18% Owner
标准 Bittensor 分配:约 41% 矿工 | 约 41% 验证者(通过债券)| 约 18% 所有者
Market | 市场与变现

From Emissions to Organic Revenue
从排放驱动到有机收入

Phase 1

Bootstrap 启动期
High emissions attract miners. Free tier builds accuracy benchmark.
高排放吸引矿工,免费层建立准确度基准。

Phase 2

Subscriptions 订阅期
Pro $50-200/mo. Enterprise API $0.01-0.05/query. Custom $500-5k.
Pro 订阅 / 企业 API / 定制报告三级变现。

Phase 3

Flywheel 飞轮期
Historical DB monetization. "Accuracy as a Service" for other AI.
历史数据库资产化,"准确度即服务"。

Competitor 竞品 Strength 优势 Gap We Fill 我们填补的空白
Think Tanks 智库 (RAND, Stratfor) Deep expertise 深度专业 $XX,XXX/year barrier; no validation loop 价格门槛高,无验证闭环
Prediction Markets 预测市场 (Polymarket) Crowd wisdom 众包智慧 Binary only; no structured foresight 二元结果,无结构化前瞻
AI Platforms (Mirofish original) Simulation power 模拟能力强 Centralized; single method; cannot evolve 中心化,单一方法,无法进化
MiroSwarm Decentralized, validated, evolving First event-prediction subnet on Bittensor
Closing | 结语

MiroSwarm

The first Bittensor subnet that turns event prediction accuracy into a tradable, verifiable, continuously improving commodity.

第一个将事件预测准确度商品化的 Bittensor 子网。

50% Accuracy Weight 准确度权重
14-Day Long Horizon 长期窗口
Dual Track Output 双轨输出
Real World Truth 真实验证

Thank you. Questions welcome. | 感谢聆听,欢迎提问。