feat: Phase 2 — Self-improving feedback loop with per-category calibration

.github/workflows/tests.yml

@@ -0,0 +1,55 @@
+name: Tests
+
+on:
+  push:
+    branches: [main]
+  pull_request:
+    branches: [main]
+
+jobs:
+  backend-tests:
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: backend
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Set up Python
+        uses: actions/setup-python@v5
+        with:
+          python-version: "3.11"
+
+      - name: Install uv
+        uses: astral-sh/setup-uv@v4
+
+      - name: Install dependencies
+        run: uv sync --dev
+
+      - name: Run tests
+        run: uv run python -m pytest tests/ -v --tb=short
+        env:
+          LLM_API_KEY: "test-key"
+
+  frontend-build:
+    runs-on: ubuntu-latest
+    defaults:
+      run:
+        working-directory: frontend
+
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v4
+
+      - name: Set up Node.js
+        uses: actions/setup-node@v4
+        with:
+          node-version: "20"
+
+      - name: Install dependencies
+        run: npm ci
+
+      - name: Build
+        run: npm run build

.gitignore

@@ -2,7 +2,7 @@
 .DS_Store
 Thumbs.db
 
-# 环境变量（保护敏感信息）
+# Environment variables
 .env
 .env.local
 .env.*.local
@@ -36,7 +36,7 @@ yarn-error.log*
 *.swp
 *.swo
 
-# 测试
+# Test
 .pytest_cache/
 .coverage
 htmlcov/
@@ -45,17 +45,20 @@ htmlcov/
 .cursor/
 .claude/
 
-# 文档与测试程序
+# Documentation and test programs
 mydoc/
 mytest/
 
-# 日志文件
+# Log files
 backend/logs/
 *.log
 
-# 上传文件
+# Uploads
 backend/uploads/
 
-# Docker 数据
+# SQLite data
+backend/data/
+
+# Docker
 data/backend/venv311/
 backend/venv311/

DESIGN.md

@@ -0,0 +1,118 @@
+# MiroFish Design Tokens
+
+Extracted from `PredictionView.vue`. All new views and components must follow these tokens.
+
+---
+
+## Typography
+
+| Role        | Family                                            | Variable   |
+|-------------|---------------------------------------------------|------------|
+| Monospace   | `'JetBrains Mono', 'SF Mono', monospace`          | `--mono`   |
+| Sans-serif  | `'Space Grotesk', 'Noto Sans SC', system-ui, sans-serif` | `--sans` |
+
+### Sizes & Weights
+
+| Element           | Size    | Weight | Letter-spacing |
+|-------------------|---------|--------|----------------|
+| Page title        | 28px    | 700    | -0.02em        |
+| Section heading   | 16px    | 600    | -0.01em        |
+| Body text         | 14px    | 400    | normal         |
+| Small / caption   | 12px    | 500    | 0.02em         |
+| Badge label       | 11px    | 600    | 0.05em         |
+| Mono data         | 13-14px | 500    | normal         |
+
+---
+
+## Colors
+
+| Token              | Hex       | Usage                          |
+|--------------------|-----------|--------------------------------|
+| `--text-primary`   | `#000`    | Nav, headings, primary text    |
+| `--orange`         | `#FF4500` | Accent, links, active states   |
+| `--green`          | `#10B981` | Success, BUY signal, positive  |
+| `--red`            | `#dc2626` | Error, SELL signal, negative   |
+| `--amber`          | `#F59E0B` | Warning, MEDIUM tier           |
+| `--amber-bg`       | `#FFFBEB` | Amber background fill          |
+| `--border`         | `#EAEAEA` | Panel borders, dividers        |
+| `--bg-subtle`      | `#FAFAFA` | Subtle background fills        |
+| `--text-secondary` | `#666`    | Secondary labels               |
+| `--text-muted`     | `#999`    | Muted / tertiary text          |
+
+---
+
+## Spacing
+
+| Token       | Value   | Usage                      |
+|-------------|---------|----------------------------|
+| Max-width   | 1400px  | Page container             |
+| Padding     | 40px    | Container horizontal pad   |
+| Grid gap    | 30px    | Between panel columns      |
+
+---
+
+## Components
+
+### Panels
+- Border: `1px solid var(--border)`
+- Border-radius: **0** (no rounded corners)
+- Background: `#fff`
+- No box-shadow
+
+### Badges
+- Uppercase text, `11px` font, `600` weight, `0.05em` letter-spacing
+- Padding: `4px 10px`
+- Border: `1px solid` (color matches text)
+- No border-radius
+
+### Skeleton Loaders
+- Background: `var(--bg-subtle)`
+- Shimmer animation (left-to-right sweep)
+- Match the dimensions of the content they replace
+
+### Empty States
+- Centered text, muted color (`var(--text-muted)`)
+- Optional icon above text
+
+### Progress Bars
+- Track: `var(--bg-subtle)`
+- Fill: `var(--orange)` or signal color
+- Height: 4-6px
+- No border-radius
+
+---
+
+## Anti-patterns
+
+Do **not** use:
+- Rounded corners (`border-radius`)
+- Box shadows (`box-shadow`)
+- Gradient fills (`linear-gradient`, `radial-gradient`)
+
+---
+
+## CSS Variables Reference
+
+```css
+:root {
+  --mono: 'JetBrains Mono', 'SF Mono', monospace;
+  --sans: 'Space Grotesk', 'Noto Sans SC', system-ui, sans-serif;
+  --orange: #FF4500;
+  --green: #10B981;
+  --red: #dc2626;
+  --border: #EAEAEA;
+  --bg-subtle: #FAFAFA;
+  --text-primary: #000;
+  --text-secondary: #666;
+  --text-muted: #999;
+}
+```
+
+---
+
+## Responsive Breakpoints
+
+| Breakpoint | Target  | Notes                             |
+|------------|---------|-----------------------------------|
+| 1024px     | Tablet  | Stack grid to single column       |
+| 768px      | Mobile  | Reduce padding, smaller type      |

README.md

@@ -36,6 +36,8 @@ The [original MiroFish](https://github.com/666ghj/MiroFish) was built for the Ch
 3. **Simulation** — Agents interact on simulated social platforms: posting, replying, arguing, shifting opinions. The system tracks sentiment evolution, topic propagation, and influence dynamics in real time.
 4. **Report** — A ReportAgent analyzes the post-simulation environment, interviews a focus group of agents, searches the knowledge graph for evidence, and generates a structured analysis.
 5. **Interaction** — Chat with any agent from the simulated world. Ask them why they posted what they posted. Full memory and personality persists.
+6. **Prediction Markets** — Browse live Polymarket markets, run a multi-agent debate simulation, and generate calibrated trading signals (BUY_YES / BUY_NO / HOLD) with edge and confidence scores.
+7. **Backtesting** — Validate signal quality against resolved markets. Computes accuracy, Brier score, ROI, Sharpe ratio, max drawdown, and calibration RMSE. Paper trading mode simulates execution with slippage.
 
 ## Screenshot
 
@@ -137,12 +139,15 @@ This fork introduces a clean abstraction layer between the application and the g
 ┌─────────────────────────────────────────┐
 │              Flask API                   │
 │  graph.py  simulation.py  report.py     │
+│  prediction.py  backtest.py             │
 └──────────────┬──────────────────────────┘
                │ app.extensions['neo4j_storage']
 ┌──────────────▼──────────────────────────┐
 │           Service Layer                  │
 │  EntityReader  GraphToolsService         │
 │  GraphMemoryUpdater  ReportAgent         │
+│  PredictionManager  Backtester           │
+│  Calibrator  PaperTrader                 │
 └──────────────┬──────────────────────────┘
                │ storage: GraphStorage
 ┌──────────────▼──────────────────────────┐
@@ -171,6 +176,8 @@ This fork introduces a clean abstraction layer between the application and the g
 - Hybrid search: 0.7 × vector similarity + 0.3 × BM25 keyword search
 - Synchronous NER/RE extraction via local LLM (replaces Zep's async episodes)
 - All original dataclasses and LLM tools (InsightForge, Panorama, Agent Interviews) preserved
+- Prediction pipeline: market → scenario → LLM debate → calibrated signal (60-90s per market)
+- SQLite (WAL mode) for backtest results, paper trading positions, calibration models
 
 ## Hardware Requirements
 
@@ -203,3 +210,5 @@ This is a modified fork of [MiroFish](https://github.com/666ghj/MiroFish) by [66
 - Entire frontend translated from Chinese to English (20 files, 1,000+ strings)
 - All Zep references replaced with Neo4j across the UI
 - Rebranded to MiroFish Offline
+- Prediction market signal engine (Polymarket integration, LLM debate simulation)
+- Backtesting + paper trading system with SQLite storage and 62-test suite

ROADMAP.md

@@ -1,8 +1,8 @@
 # MiroFish-Offline Roadmap
 
-## Current State (v0.2.0)
+## Current State (v0.2.0+)
 
-Fully local fork running on Neo4j CE + Ollama. All Zep Cloud dependencies removed. Core pipeline works: upload text → build knowledge graph → entity extraction → simulation → report generation.
+Fully local fork running on Neo4j CE + Ollama. All Zep Cloud dependencies removed. Core pipeline works: upload text → build knowledge graph → entity extraction → simulation → report generation. Prediction market signal engine with backtesting, paper trading, and SQLite storage. 62-test suite.
 
 ---
 
@@ -52,7 +52,7 @@ Fully local fork running on Neo4j CE + Ollama. All Zep Cloud dependencies remove
 - [ ] Authentication & multi-user support
 - [ ] Graph versioning: snapshot and restore graph states
 - [ ] Plugin system for custom NER extractors, search strategies, and report templates
-- [ ] Comprehensive test suite (unit + integration + E2E)
+- [x] Comprehensive test suite — 62 tests (unit + integration) for prediction/backtest system
 - [ ] Performance benchmarks: document throughput (texts/min) and latency per hardware tier
 - [ ] Helm chart for Kubernetes deployment
 

TODOS.md

@@ -0,0 +1,12 @@
+# MiroFish TODOs
+
+## Backlog
+
+(empty)
+
+## Completed
+
+- [x] ~~**P2** JSON to SQLite migration for historical prediction runs~~ — prediction_runs table + SQLitePredictionStore + migration script
+- [x] ~~**P2** CI/CD pipeline via GitHub Actions~~ — .github/workflows/tests.yml (backend pytest + frontend build)
+- [x] ~~**P2** SQLite disk-full error handling~~ — StorageError + _safe_write wrapper on all write methods
+- [x] ~~**P2** Extract shared CSS components before Phase 2~~ — Phase 2 shipped without this; CSS components added inline per view

backend/app/__init__.py

@@ -1,12 +1,11 @@
 """
-MiroFish Backend - Flask Application Factory
+MiroFish Backend — Flask application factory
 """
 
 import os
 import warnings
 
-# Suppress multiprocessing resource_tracker warnings (from third-party libraries like transformers)
-# Must be set before all other imports
+# Suppress multiprocessing resource_tracker warnings from third-party libs
 warnings.filterwarnings("ignore", message=".*resource_tracker.*")
 
 from flask import Flask, request
@@ -17,19 +16,17 @@
 
 
 def create_app(config_class=Config):
-    """Flask application factory function"""
+    """Flask application factory"""
     app = Flask(__name__)
     app.config.from_object(config_class)
 
-    # Configure JSON encoding: ensure Chinese displays directly (not as \uXXXX)
-    # Flask >= 2.3 uses app.json.ensure_ascii, older versions use JSON_AS_ASCII config
+    # JSON encoding: display CJK characters directly (not \uXXXX)
     if hasattr(app, 'json') and hasattr(app.json, 'ensure_ascii'):
         app.json.ensure_ascii = False
 
-    # Setup logging
     logger = setup_logger('mirofish')
 
-    # Only print startup info in reloader subprocess (avoid printing twice in debug mode)
+    # Only log startup info in the reloader child process (avoid duplicate logs in debug mode)
     is_reloader_process = os.environ.get('WERKZEUG_RUN_MAIN') == 'true'
     debug_mode = app.config.get('DEBUG', False)
     should_log_startup = not debug_mode or is_reloader_process
@@ -54,19 +51,30 @@ def create_app(config_class=Config):
         # Store None so endpoints can return 503 gracefully
         app.extensions['neo4j_storage'] = None
 
-    # Register simulation process cleanup function (ensure all simulation processes terminate on server shutdown)
+    # --- Initialize SQLite storage ---
+    from .storage.sqlite_store import SQLiteStore
+    try:
+        sqlite_store = SQLiteStore(Config.SQLITE_DB_PATH)
+        app.extensions['sqlite'] = sqlite_store
+        if should_log_startup:
+            logger.info("SQLiteStore initialized (%s)", Config.SQLITE_DB_PATH)
+    except Exception as e:
+        logger.error("SQLiteStore initialization failed: %s", e)
+        app.extensions['sqlite'] = None
+
+    # Register simulation process cleanup
     from .services.simulation_runner import SimulationRunner
     SimulationRunner.register_cleanup()
     if should_log_startup:
-        logger.info("Simulation process cleanup function registered")
+        logger.info("Simulation process cleanup registered")
 
     # Request logging middleware
     @app.before_request
     def log_request():
         logger = get_logger('mirofish.request')
         logger.debug(f"Request: {request.method} {request.path}")
         if request.content_type and 'json' in request.content_type:
-            logger.debug(f"Request body: {request.get_json(silent=True)}")
+            logger.debug(f"Body: {request.get_json(silent=True)}")
 
     @app.after_request
     def log_response(response):
@@ -75,18 +83,19 @@ def log_response(response):
         return response
 
     # Register blueprints
-    from .api import graph_bp, simulation_bp, report_bp
+    from .api import graph_bp, simulation_bp, report_bp, prediction_bp, backtest_bp
     app.register_blueprint(graph_bp, url_prefix='/api/graph')
     app.register_blueprint(simulation_bp, url_prefix='/api/simulation')
     app.register_blueprint(report_bp, url_prefix='/api/report')
+    app.register_blueprint(prediction_bp, url_prefix='/api/prediction')
+    app.register_blueprint(backtest_bp, url_prefix='/api/backtest')
 
     # Health check
     @app.route('/health')
     def health():
         return {'status': 'ok', 'service': 'MiroFish-Offline Backend'}
 
     if should_log_startup:
-        logger.info("MiroFish-Offline Backend startup complete")
+        logger.info("MiroFish-Offline Backend started")
 
     return app
-