astra-quant-agent

🚀 AstraQuant v8.6.0: Regime-Adaptive Dual-Mode Engine, Anti-Chop Protocol & Agile Capital Management

AstraQuant v8.6.0 Official Feature Release is live! This major release addresses a critical real-world trading dilemma: protecting capital during choppy, range-bound markets while maximizing trend capture. It introduces the Regime-Adaptive Dual-Mode Architecture, an uncompromising Anti-Chop Protocol, agile breakeven ratcheting, and tightened capital rotation.

⚡ 1. Regime-Adaptive Dual-Mode Engine (Trend vs. Chop)

• Autonomous Market Regime Classifier: Dynamic instrument-level classification combining 4H macro channel, 1H swing structure, and 1H ADX:
  • STRONG_TREND: 4H macro and 1H swing aligned + 1H ADX ≥ 20.0 (aggressive pullback buying, trend breakouts, distant multi-tier TP2 trailing).
  • CHOP_RANGE: 4H range, 1H chop, or 1H ADX < 20.0 (strict defensive posture, zero momentum chasing, boundary mean-reversion only).
• Prompt Architecture Regime Injection: Real-time regime diagnosis banners injected directly into each asset's triple-barrier structure (- 🏛️ 三重滤网宏观结构: ... | 标的体制=CHOP_RANGE (区间震荡·严禁追单)), providing LLM CIO agents with explicit, unambiguous market context.

🛡️ 2. Quantitative Anti-Chop Protocol & Signal Engine Hardening

• Suppression of False Breakouts: Programmatically disables Setup 3 (🚀 Momentum Breakout) and Setup 4 (🌪️ Breakdown Acceleration) when ADX indicates range-bound conditions, eliminating the fatal trap of buying range tops and shorting range bottoms.
• Extreme Boundary Mean Reversion: Restricts chop entries exclusively to Setup 5 (💎 Extreme Mean Reversion) and Setup 6 (🛡️ Sweep Reversal) at 24H VWAP Value Area extremes (VAL / VAH) with confirmed exhaustion wicks.
• Mid-Range Noise Gating: Zeroes out deceptive temporary EMA alignment scores in chop; enforces strict HOLD when an asset lacks extreme reversal setups, preventing random sub-factor sum bleed.
• Elevated Confidence Threshold: Increases minimum entry confidence to ≥ 82% in chop regimes.

🎯 3. Agile Breakeven Ratchet & Tightened Capital Rotation

• Chop-Adaptive Breakeven Trigger:
  • In normal/trend markets: preserves the standard 1.5x ATR trigger for wave breathing.
  • In chop regimes: ratchets breakeven stop loss forward at 0.9x ATR peak profit (~0.6R–0.7R), locking in cost + 0.2% fee cushion before oscillating swings reverse.
• Compact Volatility Time-Stop: Shortens time-stop from 4.0 hours to 2.5 hours (0.625x) for range-bound positions, immediately freeing up collateral and avoiding dead capital lockup.
• Near-End Structural Take-Profit: Prohibits unreachable distant targets (>2.0x ATR) in chop; mandates pinning TP to range centers (VWAP / POC) or opposite bounds.

🎨 4. Workstation & Landing Page Upgrades

• Landing Page Modernization: Refreshed visual aesthetics, high-conversion typography, and interactive trading preview card.
• Exchange Referral Channels: Integrated official fee rebate channels for OKX, Binance, and Gate.
• Standardized Community Templates: Added GitHub Issue & Pull Request templates (.github/).

🧪 5. Multi-Tier Quality Gate Verification

• Tier 0 (Core Quant Gates): 1,545 Python tests + 459 frontend tests passing 100%.
• Tier 1 (Multi-Model & UI Gates): 3,909 Python tests passing + Vue production bundle compiled + vue-tsc type-check clean.
• Tier 2 (Full System & Audit Gates): 2,531 architecture and invariant audit tests passing 100%.

📦 Upgrade Guide

git pull origin main
# For Docker deployments:
./deploy/docker-start.sh
# For Bare Metal deployments:
sh deploy/install.sh && ./start.sh

🚀 AstraQuant v8.5.1: Clean Text Notification Engine, Prompt Engineering Guide Overhaul & Risk Guardrail Alignment

AstraQuant v8.5.1 Maintenance & Quality Release is officially released! This release focuses on de-slopping the notification engine, overhauling the quantitative prompt authoring guide, hardening risk parameters, and ensuring seamless custom AI model configuration.

📢 1. Notification Engine Upgrade & Quantitative Text De-Slop

• Emoji & Noise Removal: Stripped noisy and emotional emojis (🟢, 🔴, 🚀, 🎉, 🏷️) across trading notifications, replacing them with institutional bracketed labels ([实盘开仓], [限价成交], [风控止损], [阶梯止盈 TP1 达成]).
• Enriched Quantitative Metadata: Open alerts now explicitly report the active AI decision model (AI主脑：{active_model}) and order execution mode (委托单型：市价单/限价单), alongside 1H ATR volatility references.
• Physical Message Sanitizer: Hardened gateway worker with an automated HTML/SVG/IMG sanitizer (sanitize_message_content), stripping rogue rich-text and SVG payloads to guarantee clean display across notification channels.

📖 2. Complete Rewrite of the Prompt Engineering Guide (docs/PROMPT_GUIDE.md)

• Full Alignment with v8.5+ Architecture: Purged historical deprecated calculus concepts (v/a/j/I/E/A/VaR) and fake-zero macro regime fallbacks.
• 7-Tier Microstructure Factor Reference: Standardized prompt integration for T0 Derivatives, T0.5 Orderflow CVD, T1 Orderbook OBI, T1.5 Options Skew, T2 Term Basis, T3 VWAP/POC, and T4 Normalized MACD.
• 5 Golden Rules for High-Winrate Swings:
  1. Structural Breathing: Refrain from premature breakeven trailing; grant 1.8x–2.2x ATR breathing room.
  2. Let Runners Run: Eliminate panic market exits during healthy intraday retracements.
  3. Directional Beta Discipline: Enforce higher entry confidence thresholds (≥85%) when holding ≥2 correlated positions.
  4. Post-Stop Composure: Programmatic cooldown enforcement against revenge trading.
  5. Maker Limit Geometry: Strict geometric enforcement with positive mathematical expectancy (R:R ≥ 1.6).
• Prompt Studio Variable Dictionary: Synchronized in-app metadata in scripts/prompt_library.py to match the 7-tier factor matrix.

🛡️ 3. Execution Risk Guardrail & LLM Configuration Hardening

• Position Cap Invariant Alignment: Corrected cross-field validation between concurrent position cap and same-direction limits, ensuring long-running backend processes reload risk snapshots without drift (stale: false).
• Custom LLM Provider Dual-Sync: Streamlined quick onboarding modal to pure custom provider inputs (Base URL, Model Name, API Key) with atomic dual-synchronization to data/llm_models.json.

🧪 4. Full Quality Gate Verification

• Tier 0 (Core Quant Gates): 1,538 Python tests + 459 frontend tests passing 100%.
• Tier 1 (Multi-Model & UI Gates): 3,904 Python tests passing + Vue production bundle compiled + vue-tsc type-check clean.
• Tier 2 (Audit & Invariants): 3,751 tests passing 100%.

📦 Upgrade Guide

git pull origin main
# For Docker deployments:
./deploy/docker-start.sh
# For Bare Metal deployments:
sh deploy/install.sh && ./start.sh

🚀 AstraQuant v8.5.0: Autonomous Quant OS, Prompt Caching Architecture & Ethereal Workstation

AstraQuant v8.5.0 Official Release is here! This milestone release marks the full maturation of our autonomous quant trading operating system, delivering enterprise-grade Prompt Caching & Session Affinity, high-end Ethereal Glassmorphic UI, GNU AGPL-3.0 + Commons Clause v1.0 + Anti-Scam Rider licensing, a 2-minute Zero-Config Setup Wizard, and a comprehensive overhaul of our documentation center.

⚡ 1. Upstream Prompt Caching & Session Affinity Architecture (2026-10)

Multi-agent quant desks usually suffer from massive LLM token overheads and latency. v8.5.0 introduces an institutional-grade prompt caching architecture:

• Shared Market & Rules Prefix (>90% cached): Static trading doctrine, risk mandates, and the 7-tier quant factor matrix are consolidated into the prefix, enabling multi-agent council seats (Trend, Momentum, Quant, Macro) to hit the same KV cache block.
• Monotonic Volatility Hierarchy: Strict structuring by mutation frequency:
  • Zone 0 (Immutable): Core doctrine, role identities, and read-only JSON Schema.
  • Zone 1 (Slow-moving): 6-hour self-evolution memory lessons.
  • Zone 2 (Medium-moving): 10-minute news intelligence and sentiment scores.
  • Zone 3 (Cycle-moving): 15-minute 7-tier quant factor matrix.
  • Zone 4 (Fast-moving): Second-level orderbook ticks and millisecond timestamps.
• Claude Ephemeral Breakpoints: Native injection of cache_control: {"type": "ephemeral"} at Zone boundaries; 64/128-token chunk alignment for OpenAI and DeepSeek models.
• Float Anti-Jitter: Mathematical rounding stabilization prevents minute decimal fluctuations from busting upstream cache keys.
• L1 In-Memory + SQLite Persistent Cache: Eliminates redundant remote roundtrips during council debate and failover sequences.

🎨 2. Ethereal Glassmorphism & High-End Minimalist UI

• Ethereal Glass Surfaces: Translucent frosted glass layering (rgba(13, 15, 21, 0.76) + backdrop-filter: blur(20px) saturate(180%)) with subtle ambient spatial depth.
• Dual-Bezel Hairline Inset Highlights: Ultra-fine chamfered borders (inset 0 1px 0 rgba(255, 255, 255, 0.09)) replacing heavy borders with crisp optical edges.
• Tactile Micro-Interactions: Custom buttons with active depression kinematics, polished focus rings, and zero cyberpunk visual pollution.
• Sanitized Framework Comments: Completely cleaned all legacy external platform attribution comments across CSS and Vue components for pristine production delivery.

⚖️ 3. GNU AGPL-3.0 + Commons Clause v1.0 + Anti-Scam Rider

To protect open-source quant contributors and community users from commercial resale, predatory closed-sourcing, and financial scams:

• Strong Copyleft for Network Services (AGPL-3.0): Anyone modifying and hosting the system over a network must make full source code available to users.
• No Commercial Resale or Paid Signals (Commons Clause v1.0): Prohibits selling, renting, or charging subscription fees for the software or derived signal channels. Private self-hosted trading and academic research remain 100% free.
• Anti-Scam Rider: Strictly bans fraudulent financial schemes, token issuance, and unauthorized paid copy-trading groups.

🚀 4. Production Deployment & Onboarding Overhaul

• 2-Minute Interactive Setup Wizard (./setup.sh): Fast CLI onboarding for exchange mode (Demo/Live), API keys, AI model provider, risk baseline, and secure password generation; supports --non-interactive for automated server deployments.
• Docker Compose Clean-up: Completely eliminated stale ./plugins mounts following the 2026-10 strategy interceptor retirement, preventing root-owned host folder pollution.
• Adaptive Execution: install.sh now upgrades pip automatically before installing dependencies; systemd services properly inherit virtualenv binar

🚀 AstraQuant v8.4.0: Obsidian-Emerald Workstation, Dynamic Sizing & Production Engine

AstraQuant v8.4.0 is a milestone release delivering an all-new Obsidian-Emerald visual brand identity, dynamic equity-proportional sizing, high-frequency prompt & risk suites, a live-only Strategy Plaza sharing endpoint, unified 3-source error center, and best-in-class bilingual documentation with updated high-resolution telemetry screenshots.

🌟 What's New in v8.4.0

1. 🖥️ Obsidian-Emerald Trading Workstation & Brand Upgrade

• Visual Identity: Upgraded to the new institutional Obsidian-Emerald design system (#00E599 emerald accent on deep obsidian slate) with the Concept C Astral Spark brand logo.
• KLineChart v10 Native Candlestick Engine: High-fidelity candlestick chart rendering live BBO maker limit entry, multi-target TP1/TP2 brackets, trailing stop lines, and depth metrics with zero horizontal line truncation or mobile indicator clipping.
• Brand-New Landing Page: Introduced a modern, responsive showcase page at / with real-time connectivity telemetry, features breakdown, and instant access to the trading desk.

2. ⚡ Dynamic Margin Sizing & Capital Allocation

• Pure Dynamic Equity Sizing: Completely eliminated historical fixed USD floors. All 1R risk budgets, single-trade margin bounds, cumulative account limits, and daily drawdown halts are now purely proportional to account equity ($1R = \text{min}(\text{cap}, \text{equity} \times 2.0%)$).
• Proportional Scaling: Seamless scaling from a 20 USDT demo trial to 10,000+ USDT institutional capital pools without configuration hacks.
• Order Execution Mode: Prioritized order mode selection in /admin/security, supporting both BBO Maker limit execution and instant Market orders.

3. 👥 Multi-Model Investment Committee & Fail-Closed Risk Pipeline

• Heterogeneous LLM Routing: Each committee seat independently binds to its own provider, model, and reasoning temperature (Claude 3.7 / DeepSeek-R1 / GPT-4o / Gemini).
• Thinking Budget Control: Global long-CoT reasoning budget support (10s to 1800s) with millisecond-grade failover across backup model chains on HTTP 429 / provider outages.
• 4 Factory Interceptor Gates: Non-bypassable physical Python risk pipeline (4H macro trend veto, entry confidence gate, 1H ADX chop filter, and 2.0R risk/reward floor). Fail-closed by design: any error or timeout unconditionally halts order entry.

4. 🌐 Three-Venue Equal Parity & Strategy Plaza

• OKX, Binance & Gate Parity: Equal status across all 3 venues with unified asset telemetry, native hedge-mode handling, and atomic conditional orders.
• Strategy Plaza Live Sharing: Introduced /api/v1/public/plaza/profile telemetry and one-click clone endpoint with a strict live-trading preflight guard (rejects demo mode) and physical zero-secret sanitization.
• Admin Control: Granular privacy switches to conceal exact equity, trades history, or active positions while sharing strategy fingerprints.

5. 🧬 Closed-Loop Self-Evolution & 3-Source Error Center

• 6-Hour Heuristic Evolution: Mines closed-trade ledgers for win rate, profit factor, and attribution slices, distilling operational wisdom into long-term prompt memory with a 7–14 day sharpness half-life.
• Consolidated Error Center: Unified system logs under /admin/decisions with a dedicated 3-source Error Center aggregating trader daemon, FastAPI backend, and gateway scheduler errors.
• Full Lifecycle Untruncated Ledger: Delivered untruncated trade lifecycle history with pagination support and all-time statistics.

6. 📖 Best-in-Class Documentation & Updated Visuals

• Comprehensive Bilingual Documentation: Completely rewritten README.md (English) and README.zh-CN.md (Chinese) covering core architecture, mathematical calculus dynamics, risk mechanics, and developer code maps.
• 15 Fresh Screenshots: Replaced legacy grap