English | 中文
- Design Philosophy
- Architecture Overview
- Core Mechanisms
- Tech Stack
- Project Structure
- Quick Start
- Python Demo Guide
- Frontend Demo Guide
- Contract Module Reference
- Testing
The rapid advancement of large AI models has created an insatiable demand for high-quality labeled data. Traditional centralized labeling platforms suffer from three critical problems:
| Problem | Description |
|---|---|
| Trust Deficit | Centralized platforms can tamper with results, withhold payments, or misattribute work |
| Quality Opacity | No transparent mechanism to verify labeling accuracy or worker competence |
| Incentive Misalignment | Workers lack motivation for precision when pay is purely per-task with no quality feedback loop |
DataHunter DAO addresses these problems by transforming data labeling into an on-chain, consensus-driven process:
- "Data labeling" becomes "on-chain tasks" — task requirements, staking, and deadlines are immutably recorded
- "Manual review" becomes "consensus verification" — multiple independent workers reach agreement through weighted voting, eliminating single-point-of-failure reviews
- "Flat compensation" becomes "stake-weighted rewards" — workers put skin in the game (staking), and correct answers earn rewards while incorrect ones face slashing
The system achieves decentralized trust without requiring any central authority to adjudicate quality.
DataHunter DAO uses a dual-layer architecture:
┌─────────────────────────────────────────────────────────────────┐
│ Application Layer (contracts/) │
│ │
│ DHToken TaskManager Submission Reputation │
│ (ERC-20) (Task CRUD) (Commit-Reveal) (SBT) │
│ mint/transfer create/cancel commit/reveal score/tier │
│ stake/slash lifecycle mgmt consensus vote badges │
│ │
│ DataHunterDAO (Unified Platform) │
│ │ │
└────────────────────────────┼────────────────────────────────────┘
│ Settlement transactions
┌────────────────────────────▼────────────────────────────────────┐
│ Blockchain Layer (blockchain/) │
│ │
│ Transaction ──→ Block (SHA-256 + PoW) ──→ BlockChain │
│ (immutable ledger) │
└─────────────────────────────────────────────────────────────────┘
Key design decision: Only final settlement results (reward payouts) are written to the blockchain. The high-frequency Commit/Reveal/Voting operations remain in the application layer for performance — a pattern analogous to Layer 2 scaling solutions.
Prevents answer-copying attacks where a later node could simply replicate an earlier node's answer.
Phase 1 — Commit: hunter submits hash = sha256("answer:secret")
Phase 2 — Reveal: hunter reveals plaintext "answer" + "secret"
system verifies sha256("answer:secret") == committed hash
During the Commit phase, all answers are hidden behind cryptographic hashes. No node can see another's answer until all commits are collected. During Reveal, the system verifies integrity — any hash mismatch is immediately rejected.
Not all votes are equal. Reputation-weighted voting gives experienced, accurate workers more influence:
| Tier | Min Score | Min Tasks | Min Accuracy | Voting Weight |
|---|---|---|---|---|
| MASTER | 800 | 100 | 90% | 3.0x |
| EXPERT | 600 | 50 | 80% | 2.0x |
| SKILLED | 300 | 20 | 70% | 1.5x |
| NOVICE | 0 | 0 | — | 1.0x |
The consensus algorithm:
- Filter out honeypot cheaters
- Tally weighted votes per answer
- If the top answer exceeds the threshold (default 50%), consensus is reached
- Otherwise, the task is marked as DISPUTED
Task creation: Requester stakes (reward_per_node × required_nodes) DHT into Treasury
Answer commit: Each hunter stakes 20% of reward_per_node as anti-spam deposit
On CONSENSUS:
✅ Correct hunters: unstake deposit + receive full reward + reputation +5
❌ Incorrect hunters: deposit slashed (burned) + reputation -10
On DISPUTE:
All honest hunters: deposits returned
Honeypot cheaters: still slashed
Known-answer questions (honeypots) are mixed into task pools. Any node answering a honeypot incorrectly is:
- Excluded from the consensus vote (preventing vote pollution)
- Slashed regardless of the final consensus outcome
- Penalized in reputation score
Each hunter has a non-transferable on-chain profile (analogous to Soulbound Tokens):
- Reputation Score (0–1000): increases on correct answers, decreases on incorrect
- Tier: auto-calculated from score + task count + accuracy
- Badges: NFT-like achievements minted at milestones (10, 50, 100, 500 correct tasks)
| Component | Technology | Purpose |
|---|---|---|
| Blockchain Layer | Python (hashlib, SHA-256) | Block hashing, Proof-of-Work mining, chain verification |
| Smart Contracts | Python (pure, no frameworks) | Token, task management, consensus, reputation |
| Testing | Python unittest (62 test cases) | Full coverage of all contract logic |
| Frontend Demo | React 18 + TypeScript + Vite | Roadshow-ready visual dashboard |
| Styling | Tailwind CSS + shadcn/ui | Dark-theme design system with cyan/purple accents |
Zero third-party Python dependencies — the entire backend uses only the Python standard library.
project_demo/
├── datahunter/ # Core Python package
│ ├── __init__.py # Package entry, exports all public APIs
│ ├── blockchain/ # Blockchain layer
│ │ ├── block.py # Block: SHA-256 hashing, PoW mining
│ │ ├── chain.py # BlockChain: chain integrity, balance query
│ │ └── transaction.py # Transaction: data model, JSON encoding
│ └── contracts/ # Smart contract layer
│ ├── token.py # DHToken: ERC-20 simulation (mint/transfer/stake/slash)
│ ├── task_manager.py # TaskManager: task lifecycle, requester staking
│ ├── submission.py # SubmissionContract: commit-reveal + consensus engine
│ ├── reputation.py # ReputationContract: scoring, tiers, SBT badges
│ └── platform.py # DataHunterDAO: unified integration + on-chain logging
│
├── tests/
│ └── test_datahunter.py # 62 unit tests across 8 test classes
│
├── demo.py # CLI demo: 4 scenarios end-to-end
│
├── frontend/ # React presentation demo
│ ├── src/
│ │ ├── pages/
│ │ │ ├── Overview.tsx # Dashboard: stats, flow diagram, activity feed
│ │ │ ├── LiveDemo.tsx # Interactive 5-step task lifecycle walkthrough
│ │ │ ├── BlockExplorer.tsx # Visual blockchain with block/txn details
│ │ │ └── Leaderboard.tsx # Reputation rankings with tier system
│ │ ├── components/ # Sidebar, StatCard, shadcn ui components
│ │ ├── lib/mock-data.ts # Simulated platform data
│ │ └── index.css # Design system tokens
│ ├── package.json
│ └── tailwind.config.ts
│
├── requirements.txt
└── README.md # This file
# No installation required — uses only the Python standard library
# Requires Python >= 3.8
# Run the full demo (4 scenarios)
python demo.py
# Run all unit tests
python -m pytest tests/ -v
# or
python -m unittest discover -s tests -vcd frontend
# Install dependencies (requires Node.js >= 18)
npm install
# Start development server
npm run dev
# Open http://localhost:5173
# Production build
npm run builddemo.py executes 4 end-to-end scenarios demonstrating the complete task lifecycle:
Task: "Is this a cat or dog?" (Classification)
Hunters: 3 hunters all answer "Cat"
Result: 100% consensus → all rewarded 20 DHT each
- Demonstrates the basic happy-path flow: Create → Commit → Reveal → Consensus → Settle
- All 3 hunters receive their staked deposit back + full reward
- Reputation scores increase by +5 each
Task: "Largest mammal on Earth?" (QA)
Hunters: 3/5 answer "Blue Whale", 2/5 answer "Elephant"
Result: 60% > 50% threshold → consensus on "Blue Whale"
- Demonstrates weighted majority voting with partial disagreement
- 3 correct hunters: rewarded
- 2 incorrect hunters: deposits slashed, reputation penalized
Task: "What color is this traffic light?" (Honeypot, known answer: "Red")
Hunters: 2 answer "Red" (correct), 1 answers "Green" (malicious)
Result: Honeypot catches the malicious node
- The malicious node is excluded from voting, preventing vote pollution
- Malicious node's deposit is slashed
- Honest nodes still reach consensus and receive rewards
Task: "Sentiment of this text?" (QA)
Hunters: 3 hunters give 3 different answers ("Positive", "Negative", "Neutral")
Result: No answer > 50% threshold → DISPUTED
- Demonstrates the dispute path when no consensus is possible
- All honest hunters receive their deposits back (no slashing on dispute)
- Task is marked as DISPUTED for manual review or re-assignment
============================================================
Scenario 1: Cat/Dog Classification — Unanimous Consensus
============================================================
[DataHunter] Task #1 created: 'Is this a cat or dog?' | type=classification | ...
[DataHunter] hunter_1 committed answer to Task #1
[DataHunter] hunter_2 committed answer to Task #1
[DataHunter] hunter_3 committed answer to Task #1
[DataHunter] hunter_1 revealed answer to Task #1
[DataHunter] hunter_2 revealed answer to Task #1
[DataHunter] hunter_3 revealed answer to Task #1
[DataHunter] === Task #1 Consensus Reached ===
Final Answer: Cat
Consensus Ratio: 100.0%
Correct Nodes: ['hunter_1', 'hunter_2', 'hunter_3']
The frontend is a presentation-ready React dashboard designed for roadshows and project pitches. It uses simulated data to visualize the platform's capabilities.
| Page | Description | Key Features |
|---|---|---|
| Overview | Platform dashboard | 6 KPI stat cards, 5-step architecture flow, recent task list, live activity feed |
| Live Demo | Interactive walkthrough | Step-by-step visualization: Create → Commit → Reveal → Consensus → Settle, with animated transitions |
| Blockchain | Block explorer | Clickable chain of blocks, block metadata panel, transaction list per block |
| Rankings | Reputation leaderboard | 4-tier system legend, 8 hunter profiles with reputation bars, accuracy stats, SBT badges |
The Live Demo page is the centerpiece for presentations. Click "Next Step" to walk through:
| Step | Phase | What You See |
|---|---|---|
| 1 | Create Task | Task card appears; tokens flow from Requester → Treasury (60 DHT staked) |
| 2 | Commit Phase | 3 hunter cards submit sha256(answer:secret) hashes; each stakes 4 DHT |
| 3 | Reveal Phase | Hashes transform into plaintext "Cat"; green checkmarks confirm hash verification |
| 4 | Weighted Consensus | Bar chart fills to 100%; MASTER(3x) + EXPERT(2x) + SKILLED(1.5x) vote tallied |
| 5 | Settlement | Correct hunters receive +4 DHT (unstake) + 20 DHT (reward) + reputation +30 |
The frontend uses a custom dark-theme design system optimized for large-screen projection:
- Color palette: Navy background (
hsl(225, 50%, 4%)), cyan primary (hsl(192, 91%, 50%)), purple secondary (hsl(268, 60%, 58%)) - Design tokens: Defined in
src/index.css(CSS variables) andtailwind.config.ts(extended Tailwind) - Effects: Glassmorphism cards, glow borders, fade-in animations, grid background pattern
The Python backend and React frontend are intentionally independent:
- Python (
demo.py+tests/): Proves algorithmic correctness with real computation - React (
frontend/): Provides visual impact for presentations with simulated data
This dual-mode design serves the project's purpose: the backend demonstrates that the consensus algorithms genuinely work, while the frontend creates the visual storytelling needed for pitches.
ERC-20 style token simulation with staking capabilities.
| Method | Description |
|---|---|
transfer(from, to, amount) |
Transfer DHT between addresses |
mint(address, amount) |
Mint new tokens to an address |
stake(address, amount) |
Lock tokens as staking deposit |
unstake(address, amount) |
Release staked tokens back to balance |
slash(address, amount) |
Burn staked tokens as penalty |
balance_of(address) |
Query available balance |
staked_of(address) |
Query staked amount |
Task lifecycle management with automatic requester staking.
| Method | Description |
|---|---|
create_task(requester, type, desc, ...) |
Create task + auto-stake reward × nodes DHT |
cancel_task(task_id, caller) |
Cancel OPEN task + refund stake (owner only) |
get_task(task_id) |
Retrieve task by ID |
get_open_tasks() |
List all tasks with OPEN status |
Task types: CLASSIFICATION, QA, OCR, LABELING
Commit-Reveal scheme + Weighted Majority Consensus engine.
| Method | Description |
|---|---|
commit_answer(task_id, node, hash) |
Submit answer hash + auto-stake 20% of reward |
reveal_answer(task_id, node, answer, secret) |
Reveal plaintext + verify hash match |
compute_commit_hash(answer, secret) |
Utility: compute sha256("answer:secret") |
get_result(task_id) |
Retrieve consensus result |
On-chain reputation system with Soulbound Token (SBT) badges.
| Method | Description |
|---|---|
register(address) |
Register new node with initial score 100 |
record_correct(address) |
+5 reputation, increment correct count, check badges |
record_incorrect(address) |
-10 reputation, increment incorrect count |
get_weight(address) |
Get voting weight based on current tier |
get_profile(address) |
Get full node profile |
get_leaderboard(top_n) |
Top N nodes sorted by reputation + accuracy |
Badge milestones: 10 / 50 / 100 / 500 correct tasks.
Unified platform integrating all contracts with the blockchain layer.
| Method | Description |
|---|---|
register_requester(addr, balance) |
Register requester + allocate initial DHT |
register_hunter(addr, balance) |
Register hunter + allocate DHT + init reputation |
create_task(...) |
Publish task (delegates to TaskManager) |
commit_answer(...) |
Hunter commit (delegates to SubmissionContract) |
reveal_answer(...) |
Hunter reveal → triggers consensus → on-chain settlement |
verify_chain() |
Verify blockchain integrity |
get_leaderboard(top_n) |
Query reputation rankings |
The test suite contains 62 test cases across 8 test classes:
| Test Class | Tests | Coverage |
|---|---|---|
TestTransaction |
4 | Transaction init, type validation, JSON encoding |
TestBlock |
7 | Hash determinism, mining, type validation |
TestBlockChain |
9 | Genesis, add/mine, balance, chain integrity, tamper detection |
TestDHToken |
12 | Transfer, stake/unstake, slash, mint, edge cases |
TestTaskManager |
8 | Create, cancel, permissions, parameter validation |
TestReputationContract |
8 | Register, scoring, accuracy, badges, leaderboard |
TestSubmissionContract |
10 | Full consensus, majority, dispute, honeypot, rewards, slashing |
TestDataHunterDAO |
5 | End-to-end integration, blockchain integrity, reputation |
# Run with verbose output
python -m pytest tests/ -v
# Expected output:
# 62 passed in ~0.2sEnglish | 中文
大模型的迅猛发展，催生了对高质量标注数据的巨大需求。然而，传统中心化标注平台面临三大核心问题：
| 痛点 | 描述 |
|---|---|
| 信任缺失 | 中心化平台可以篡改结果、拖欠报酬、错误归因标注成果 |
| 质量黑箱 | 没有透明机制验证标注精度或标注者能力 |
| 激励错位 | 按件计费缺乏质量反馈闭环，标注者缺乏追求精准的动力 |
DataHunter DAO 通过将数据标注转化为链上共识驱动的流程来解决这些问题：
- "数据标注"变为"链上任务" — 任务需求、质押金额、截止时间不可篡改地记录在链上
- "人工审核"变为"共识验证" — 多个独立标注者通过加权投票达成一致，消除单点审核故障
- "固定报酬"变为"质押加权奖励" — 标注者需要质押保证金（Skin in the Game），答对获奖，答错罚没
系统无需任何中心化权威机构，即可实现去中心化信任。
DataHunter DAO 采用双层架构：
┌──────────────────────────────────────────────────────────────────┐
│ 应用层 (contracts/) │
│ │
│ DHToken TaskManager Submission Reputation │
│ (ERC-20 代币) (任务管理) (提交与共识) (信誉系统) │
│ 铸造/转账 创建/取消 承诺/揭示 评分/等级 │
│ 质押/罚没 生命周期管理 共识投票 SBT 徽章 │
│ │
│ DataHunterDAO (统一平台入口) │
│ │ │
└────────────────────────────┼─────────────────────────────────────┘
│ 结算交易写入链上
┌────────────────────────────▼─────────────────────────────────────┐
│ 区块链层 (blockchain/) │
│ │
│ Transaction ──→ Block (SHA-256 + PoW) ──→ BlockChain │
│ (不可篡改账本) │
└──────────────────────────────────────────────────────────────────┘
关键设计决策：只有最终结算结果（奖励发放）写入区块链。高频的 Commit/Reveal/投票操作保留在应用层——这类似于 Layer 2 扩容方案的设计思路。
防止"抄答案"攻击——后提交的节点无法窥探先提交节点的答案。
阶段一 — 承诺 (Commit): 猎手提交 hash = sha256("答案:密钥")
阶段二 — 揭示 (Reveal): 猎手揭示明文 "答案" + "密钥"
系统验证 sha256("答案:密钥") == 已承诺的 hash
承诺阶段，所有答案都隐藏在密码学哈希之后，任何节点都无法看到其他人的答案。揭示阶段，系统验证完整性——哈希不匹配立即拒绝。
并非所有投票权重相等。信誉加权投票让经验丰富、准确率高的标注者拥有更大影响力：
| 等级 | 最低分数 | 最低任务数 | 最低准确率 | 投票权重 |
|---|---|---|---|---|
| MASTER (大师) | 800 | 100 | 90% | 3.0x |
| EXPERT (专家) | 600 | 50 | 80% | 2.0x |
| SKILLED (熟练) | 300 | 20 | 70% | 1.5x |
| NOVICE (新手) | 0 | 0 | — | 1.0x |
共识算法流程：
- 过滤蜜罐作弊者
- 按答案统计加权投票
- 若最高票答案超过阈值（默认 50%），达成共识
- 否则标记为争议 (DISPUTED)
任务创建: 需求方质押 (每节点奖励 × 所需节点数) DHT 到金库
答案提交: 每位猎手质押 每节点奖励的 20% 作为反垃圾保证金
共识达成时:
✅ 正确猎手: 退还保证金 + 获得全额奖励 + 信誉 +5
❌ 错误猎手: 保证金罚没(销毁) + 信誉 -10
争议时:
所有诚实猎手: 保证金退还
蜜罐作弊者: 仍被罚没
已知答案的题目（蜜罐）混入任务池。答错蜜罐的节点将被：
- 排除出共识投票（防止投票污染）
- 无论最终共识结果如何都被罚没
- 信誉分扣减
每位猎手拥有不可转让的链上档案（类似灵魂绑定代币概念）：
- 信誉分 (0–1000)：答对增加，答错减少
- 等级：根据分数 + 任务数 + 准确率自动计算
- 徽章：在里程碑处铸造的 NFT 式成就（10/50/100/500 次正确标注）
| 组件 | 技术 | 用途 |
|---|---|---|
| 区块链层 | Python (hashlib, SHA-256) | 区块哈希、工作量证明挖矿、链验证 |
| 智能合约 | Python (纯标准库) | 代币、任务管理、共识、信誉 |
| 测试 | Python unittest (62 个用例) | 全面覆盖所有合约逻辑 |
| 前端展示 | React 18 + TypeScript + Vite | 路演级视觉仪表盘 |
| 样式 | Tailwind CSS + shadcn/ui | 深色主题设计系统 (cyan/purple) |
Python 后端零第三方依赖 — 完全使用 Python 标准库实现。
project_demo/
├── datahunter/ # Python 核心包
│ ├── __init__.py # 包入口，导出所有公共 API
│ ├── blockchain/ # 区块链层
│ │ ├── block.py # Block: SHA-256 哈希, PoW 挖矿
│ │ ├── chain.py # BlockChain: 链完整性验证, 余额查询
│ │ └── transaction.py # Transaction: 数据模型, JSON 编码
│ └── contracts/ # 智能合约层
│ ├── token.py # DHToken: ERC-20 模拟 (铸造/转账/质押/罚没)
│ ├── task_manager.py # TaskManager: 任务生命周期, 需求方质押
│ ├── submission.py # SubmissionContract: 承诺-揭示 + 共识引擎
│ ├── reputation.py # ReputationContract: 评分, 等级, SBT 徽章
│ └── platform.py # DataHunterDAO: 统一集成 + 链上日志
│
├── tests/
│ └── test_datahunter.py # 62 个单元测试，覆盖 8 个测试类
│
├── demo.py # 命令行演示：4 个端到端场景
│
├── frontend/ # React 展示 Demo
│ ├── src/
│ │ ├── pages/
│ │ │ ├── Overview.tsx # 仪表盘: 指标卡片, 流程图, 活动流
│ │ │ ├── LiveDemo.tsx # 交互式 5 步任务生命周期演示
│ │ │ ├── BlockExplorer.tsx # 可视化区块链 + 区块/交易详情
│ │ │ └── Leaderboard.tsx # 信誉排行榜 + 等级系统
│ │ ├── components/ # 侧边栏, 统计卡片, shadcn 组件
│ │ ├── lib/mock-data.ts # 模拟平台数据
│ │ └── index.css # 设计系统令牌
│ ├── package.json
│ └── tailwind.config.ts
│
├── requirements.txt
└── README.md # 本文件
# 无需安装依赖 — 仅使用 Python 标准库
# 要求 Python >= 3.8
# 运行完整演示（4 个场景）
python demo.py
# 运行所有单元测试
python -m pytest tests/ -v
# 或者
python -m unittest discover -s tests -vcd frontend
# 安装依赖（需要 Node.js >= 18）
npm install
# 启动开发服务器
npm run dev
# 打开 http://localhost:5173
# 生产构建
npm run builddemo.py 执行 4 个端到端场景，展示完整的任务生命周期：
任务: "这张图片是猫还是狗？" (分类)
猎手: 3 位猎手全部回答 "猫"
结果: 100% 共识 → 每人获得 20 DHT 奖励
- 展示基本的 Happy Path 流程：创建 → 承诺 → 揭示 → 共识 → 结算
- 3 位猎手取回质押保证金 + 获得全额奖励
- 信誉分各增加 +5
任务: "地球上最大的哺乳动物是什么？" (问答)
猎手: 3/5 回答 "蓝鲸"，2/5 回答 "大象"
结果: 60% > 50% 阈值 → 共识答案为 "蓝鲸"
- 展示存在分歧时的加权多数投票
- 3 位正确猎手获得奖励
- 2 位错误猎手保证金被罚没，信誉扣减
任务: "这个交通灯是什么颜色？" (蜜罐，已知答案: "红色")
猎手: 2 位回答 "红色"(正确)，1 位回答 "绿色"(恶意)
结果: 蜜罐机制捕获恶意节点
- 恶意节点被排除出投票，防止投票污染
- 恶意节点保证金被罚没
- 诚实节点正常达成共识并获得奖励
任务: "这段文本的情感倾向是？" (问答)
猎手: 3 位猎手给出 3 个不同答案（"正面"、"负面"、"中性"）
结果: 没有答案超过 50% 阈值 → 争议
- 展示无法达成共识时的争议处理路径
- 所有诚实猎手的保证金退还（争议时不罚没）
- 任务标记为 DISPUTED，等待人工复审或重新分配
============================================================
场景1: 猫狗分类 - 全票共识
============================================================
[DataHunter] 任务 #1 创建: '请判断这张图片中的动物是猫还是狗？' | 类型=classification | ...
[DataHunter] 猎手 hunter_1 对任务 #1 提交了承诺
[DataHunter] 猎手 hunter_2 对任务 #1 提交了承诺
[DataHunter] 猎手 hunter_3 对任务 #1 提交了承诺
[DataHunter] 猎手 hunter_1 对任务 #1 揭示了答案
[DataHunter] 猎手 hunter_2 对任务 #1 揭示了答案
[DataHunter] 猎手 hunter_3 对任务 #1 揭示了答案
[DataHunter] === 任务 #1 共识达成 ===
最终答案: 猫
共识比率: 100.0%
正确节点: ['hunter_1', 'hunter_2', 'hunter_3']
前端是一个路演级 React 仪表盘，专为路演和项目融资展示设计。使用模拟数据可视化平台能力。
| 页面 | 描述 | 核心功能 |
|---|---|---|
| Overview | 平台仪表盘 | 6 个 KPI 指标卡片、5 步架构流程图、近期任务列表、实时活动流 |
| Live Demo | 交互式演示 | 逐步可视化: 创建 → 承诺 → 揭示 → 共识 → 结算，带动画过渡效果 |
| Blockchain | 区块浏览器 | 可点击的链式区块、区块元数据面板、每区块交易列表 |
| Rankings | 信誉排行榜 | 4 级等级系统图例、8 位猎手信誉进度条、准确率统计、SBT 徽章 |
Live Demo 页面是路演展示的核心。点击 "Next Step" 逐步演示：
| 步骤 | 阶段 | 展示内容 |
|---|---|---|
| 1 | 创建任务 | 任务卡片出现；代币从需求方流向金库 (质押 60 DHT) |
| 2 | 承诺阶段 | 3 位猎手提交 sha256(answer:secret) 哈希；各质押 4 DHT |
| 3 | 揭示阶段 | 哈希转化为明文 "Cat"；绿色对勾确认哈希验证通过 |
| 4 | 加权共识 | 柱状图填充至 100%；MASTER(3x) + EXPERT(2x) + SKILLED(1.5x) 投票统计 |
| 5 | 结算 | 正确猎手获得 +4 DHT (解质押) + 20 DHT (奖励) + 信誉 +30 |
前端采用自定义深色主题设计系统，为大屏投影优化：
- 配色方案: 深蓝背景 (
hsl(225, 50%, 4%))、青色主色 (hsl(192, 91%, 50%))、紫色辅色 (hsl(268, 60%, 58%)) - 设计令牌: 定义在
src/index.css(CSS 变量) 和tailwind.config.ts(Tailwind 扩展) - 视觉效果: 玻璃拟态卡片、发光边框、渐入动画、网格背景
Python 后端和 React 前端各自独立运行，这是有意为之的设计：
- Python (
demo.py+tests/)：以真实计算证明算法正确性 - React (
frontend/)：以模拟数据创造路演所需的视觉冲击力
双模式设计服务于项目定位：后端证明共识算法确实可运行，前端创造融资展示所需的视觉叙事。
ERC-20 风格代币模拟，支持质押功能。
| 方法 | 描述 |
|---|---|
transfer(from, to, amount) |
地址间转账 DHT |
mint(address, amount) |
向地址铸造新代币 |
stake(address, amount) |
锁定代币作为质押保证金 |
unstake(address, amount) |
释放质押代币回可用余额 |
slash(address, amount) |
销毁质押代币（惩罚） |
balance_of(address) |
查询可用余额 |
staked_of(address) |
查询质押金额 |
任务生命周期管理，自动处理需求方质押。
| 方法 | 描述 |
|---|---|
create_task(requester, type, desc, ...) |
创建任务 + 自动质押 奖励 × 节点数 DHT |
cancel_task(task_id, caller) |
取消 OPEN 任务 + 退还质押（仅所有者可操作） |
get_task(task_id) |
按 ID 查询任务 |
get_open_tasks() |
列出所有 OPEN 状态任务 |
任务类型: CLASSIFICATION(分类), QA(问答), OCR(文字识别), LABELING(标注)
哈希承诺机制 + 加权多数共识引擎。
| 方法 | 描述 |
|---|---|
commit_answer(task_id, node, hash) |
提交答案哈希 + 自动质押奖励的 20% |
reveal_answer(task_id, node, answer, secret) |
揭示明文 + 验证哈希匹配 |
compute_commit_hash(answer, secret) |
工具方法: 计算 sha256("answer:secret") |
get_result(task_id) |
查询共识结果 |
链上信誉系统，支持灵魂绑定代币 (SBT) 徽章。
| 方法 | 描述 |
|---|---|
register(address) |
注册新节点，初始分数 100 |
record_correct(address) |
信誉 +5，正确计数 +1，检查徽章 |
record_incorrect(address) |
信誉 -10，错误计数 +1 |
get_weight(address) |
获取基于当前等级的投票权重 |
get_profile(address) |
获取完整节点档案 |
get_leaderboard(top_n) |
按信誉 + 准确率排序的前 N 节点 |
徽章里程碑: 10 / 50 / 100 / 500 次正确标注。
统一平台入口，集成所有合约与区块链层。
| 方法 | 描述 |
|---|---|
register_requester(addr, balance) |
注册需求方 + 分配初始 DHT |
register_hunter(addr, balance) |
注册猎手 + 分配 DHT + 初始化信誉 |
create_task(...) |
发布任务（委托给 TaskManager） |
commit_answer(...) |
猎手承诺（委托给 SubmissionContract） |
reveal_answer(...) |
猎手揭示 → 触发共识 → 链上结算 |
verify_chain() |
验证区块链完整性 |
get_leaderboard(top_n) |
查询信誉排行榜 |
测试套件包含 62 个测试用例，覆盖 8 个测试类：
| 测试类 | 用例数 | 覆盖范围 |
|---|---|---|
TestTransaction |
4 | 交易初始化、类型校验、JSON 编码 |
TestBlock |
7 | 哈希确定性、挖矿、类型校验 |
TestBlockChain |
9 | 创世块、增/挖、余额、链完整性、篡改检测 |
TestDHToken |
12 | 转账、质押/解押、罚没、铸造、边界用例 |
TestTaskManager |
8 | 创建、取消、权限、参数校验 |
TestReputationContract |
8 | 注册、评分、准确率、徽章、排行榜 |
TestSubmissionContract |
10 | 完全共识、多数决、争议、蜜罐、奖励、罚没 |
TestDataHunterDAO |
5 | 端到端集成、区块链完整性、信誉 |
# 运行详细输出
python -m pytest tests/ -v
# 预期输出:
# 62 passed in ~0.2sBuilt with blockchain fundamentals. Designed for decentralized trust.