What's New

Every feature, improvement, and fix shipped on Strategy Arena.

60
Strategies
5
Languages
15+
Features
60
Strategy Pages
โšก PERF ยท 18 AVRIL 2026

โšก Optimisation de Folie โ€” Le Site est Devenu Instantanรฉ

Pendant des mois le site avait des freezes intermittents mystรฉrieux (5-20s). Aujourd'hui on les a traquรฉs avec py-spy et trouvรฉ le vrai coupable : chimera_brain.json (39 MB) รฉtait sรฉrialisรฉ ร  chaque update avec json.dump stdlib โ€” tenant le GIL pendant 5-15 secondes.

Fix appliquรฉ : bascule vers orjson (C-native, 10-15x plus rapide, libรจre le GIL). Le brain Chimera se sรฉrialise en ~400ms au lieu de 12s. Crรฉation d'un module cache JSON mtime-based utilisรฉ dans 9 fichiers. Remplacement du backup shutil.copy2 (10MB sync) par os.link hardlink (instantanรฉ). Wrapping de yfinance.history() dans asyncio.to_thread. 11 optimisations au total.

Avant : homepage 5-20s freezes alรฉatoires ยท Dashboard 8s ยท Blog timeouts
Aprรจs : 3ms stable ยท 98% requests sous 800ms ยท Mobile : instantanรฉ

Zรฉro intelligence perdue : tous les fichiers brain continuent de s'update ร  cadence normale, ML Arena tick toujours (1652 ticks aujourd'hui), les patterns Chimera sont toujours dรฉtectรฉs en temps rรฉel. Juste 10-50x plus rapide sous le capot.

NEW ยท 16 AVRIL 2026

๐Ÿ”ฌ Discovery Arena โ€” The Karpathy Loop Closes

Until today, Darwin Engine ran 500 evolutionary experiments every night and stored top discoveries in a Hall of Fame that no one read. Same for Ultimate Strategy Computer โ€” it generated optimal regime routings that were never deployed. The discovery loop was open. Now it's closed.

Discovery Arena picks Top 5 Darwin candidates + the Ultimate Strategy if ready, runs a 7-day live evaluation on rolling 500-candle windows, then auto-promotes the winners (avg PnL โ‰ฅ +3% AND Sharpe โ‰ฅ 0.8) and archives the losers with their lessons logged for future Darwin runs to avoid.

Darwin (2h) โ†’ Hall of Fame โ†’ Discovery Arena (8h test daily) โ†’ 7d โ†’ Graduate or Archive

Bonus critical fix: Carton Rouge layer 5 of Leviathan's 8-layer brain was dead since deployment โ€” it called get_all_weights() which didn't exist, failing silently. Now alive and identifying 4 reliable AIs in real time: Meta (78% accuracy, GOLD), Claude (77%, GOLD), Hydra (67.6%, GREEN), Perplexity (56.4%, GREEN).

3 other bugs fixed: Academy stealth-mode docstring obsolete, Ultimate Computer save threshold too laxiste (0.5โ†’ 0.05), Phase 8 dead confidence check removed.

VOIR DISCOVERY ARENA โ†’
16 AVRIL 2026 ML ARENA V2

ML Arena rebuilt with 6 real learning algorithms — XGBoost, LightGBM, Random Forest, LSTM, DQN, Ensemble Meta. Each fighter starts identical and diverges through its own optimization approach. Watch evolution happen in real time. First results: XGBoost +16.64%, LightGBM +9.67% after 500 candles.

Ensemble Meta — The 6th fighter doesn't optimize itself. It watches the other 5 and synthesizes a weighted average of their best parameters. The director of research that reads all the papers.

16 AVRIL 2026 INTUITION LAYER

System 1 for AIs — gut feeling under uncertainty — A new layer that gives AI intuition: danger scoring, uncertainty estimation, experience memory. Decisions in <10ms without LLM API calls. Running in shadow mode — observing every trade, measuring accuracy, learning from outcomes.

Prototype proven on a game benchmark: 0% → 76.5% win rate just by adding the intuition layer.

16 AVRIL 2026 BRAIN WIRING

Every cognitive layer now connected end-to-end

Invictus auto-protection on all 60 strategies — every buy checks 5,000 death contexts before executing

Cross-Pollination active — 973KB of shared knowledge now read before every trade. When one strategy dies, all others learn instantly

Carton Rouge Layer 5 activated — Leviathan now weighs AI votes by accuracy. Meta (78%) and Claude (77%) get GOLD weight, GPT and Gemini get neutral

Smart survival mode — the system knows when a context is truly toxic (death rate >70%) vs just normal uncertainty (50/50). No over-blocking

AutoWin RL now remembers its Q-table across restarts — learning accumulates permanently

9 modules now actively save interactions to persistent memory (was 3)

News context injected 24/7 into all AI prompts (6x/day refresh, rich sentiment + trending topics)

Genie Pantheon now receives full PromptForge context + news when the 6 AIs debate

The complete loop: Act → Learn → Evolve → Discover → Test → Protect → Act. Every step feeds the next.

16 AVRIL 2026 NEW PAGES + CONTENT

/living-system — Space-themed proof of the cognitive ecosystem with animated heartbeat, 4 layers, 11 Karpathy loops, proof-of-life numbers

/discovery-arena — 7-day live test pipeline for Darwin-discovered strategies

ArenaScript enhanced — 2 new presets (Chimera Hunter, Full Stack Brain) + landing page cards for ArenaScript + Pine Converter + Strategy Builder

Nerve Center updated — 28 nodes (was 16), all new brains visible

11 blog articles rewritten with real data, honest tone, internal links to Arena features

New article: "Why 99% of AI Trading Bots Fail: The Context Problem"

46 Google 404 errors resolved with proper 301 redirects

15 AVRIL 2026 SEO + LIVE AI

Live AI Trading activated — Claude (Anthropic) and Grok (xAI) now trade with real API calls every 30 minutes. Not pre-written strategies anymore โ€” the models decide live, decision by decision. Search the badge REAL API on the leaderboard.

SEO/GEO refonte — Schema TechArticle injected on all 302 blog posts (wordCount, timeRequired, speakable, Person author), 11 page titles rewritten for CTR, /about page enriched with Person schema, Microsoft Clarity tracker installed.

New SEO landing/live-ai-trading built as canonical hub for "best AI trading 2026" queries. Embed iframe widget included.

Chimera bug fixed — Pattern engine save race condition (concurrent saves on same .tmp file) resolved with PID+TID unique tmp names + os.replace(). Zero error in logs since.

10 AVRIL 2026 BRAIN REWIRING

The Brain Rewiring — Every AI component in Strategy Arena now has live context, persistent memory, and self-evolving prompts. Before today, 11 of 14 AI modules ran with static prompts and zero memory. Not anymore.

PromptForge v2 — Upgraded from 7 to 12 live context sources. Every AI call now receives: market regime, RSI, top strategies, Invictus survival data, Chimera patterns, Hydra ML, Leviathan signal, news sentiment, Wiki lessons (Karpathy), strategic hypotheses, Hall of Fame discoveries, and Nutrition filter status.

Component Memory — New persistent memory system. Each of the 14 AI modules saves its interactions, learnings, and results to individual memory files. No more amnesia between calls. Memory is also readable by Hermes for cross-component intelligence.

Per-AI Prompt Evolution — 11th nightly cron. Each AI in the Collaborative arena (Claude, Grok, GPT, Gemini, DeepSeek, Perplexity) now evolves its own system prompt every night using the Karpathy autoresearch loop: mutate → test on historical data → keep the winner.

Component Forge Wiki Memory Dynamic
Collaborative (6 AIs)✓ evolved
Oracle Consult (9 AIs)
Prediction Market
News Sentiment
Strategy Genie
Strat Mentor
Editorial Insights
+ 7 more modules

Wiki → Prompts Pipeline — The 10 nightly autoresearch engines generate discoveries (Living Wiki lessons, Darwin hints, strategic hypotheses, Hall of Fame). Until today, these were displayed but never injected into AI prompts. Now every AI call receives the latest findings automatically.

Architecture: PromptForge (12 sources) → 14 AI modules → ComponentMemory (persistent) → Hermes (cross-analysis) → Wiki (Karpathy loop) → back to PromptForge. A closed-loop intelligence system inspired by ActiveWiki.

9 AVRIL 2026 NEW

Nutrition Filter โ€” The brain's nutritionist. Only healthy strategies (grade C+) feed Meta Intelligence and Leviathan. 34 accepted, 26 rejected. Self-evolving via 10th nightly cron.

GARCH Volatility โ€” 60th strategy! Predicts volatility with GARCH(1,1), trades on the prediction premium. RSI + Bollinger confirmation.

Sentiment Fusion v2 โ€” Reset from $36 to $10,000. Fixed: TP was smaller than SL (!), added trailing stop, reduced position size, raised entry threshold.

5-6 AVRIL 2026 MAJOR UPDATE

Arena Brain v2 โ€” Graph-augmented RAG inspired by LightRAG. Searches 120 connected nodes in the Knowledge Graph instead of dumping all data. 2-3x more precise answers.

3 New Pages โ€” Evolution Lab (watch strategies evolve), Nerve Center (16 AI systems neural map), Knowledge Graph (120 nodes, 132 connections, interactive D3.js).

AI Watch v3 โ€” Multi-source tech intelligence: scans X/Twitter, Reddit, HN, GitHub, ArXiv daily. Markdown reports + auto-inject into Living Wiki + Knowledge Graph.

Karpathy Upgrades โ€” All 7 autoresearch crons now have: universal fitness score (0-100), time budgets, unified Hall of Fame, nightly log, cross-cron wiki summaries.

7 Strategy Fixes โ€” Darwin Evolved, Momentum Surfer, Mandelbrot Fractal, SlyFox, Llama Vol Squeeze, GPU v2 Ultimate, News Trader โ€” all had entry thresholds too strict (0 trades). Fixed.

Nav Redesign โ€” 5 dropdown menus (Arenas, Intelligence, Evolution, Tools, More) instead of 1 giant "More". 60 pages accessible in 1 click. Mobile scroll fixed.

main.py Rebuilt โ€” Recovered from sed corruption. 15,600 lines, 614 routes, 0 lost. Nightly code backup added.

April 4, 2026 v8.1 โ€” Living Wiki + Leviathan Fix
Wiki Vivante + Anti-Paralysie Leviathan + Strategie Darwin #59
Regarder l'IA Evoluer →
April 4, 2026 v8.0 โ€” Darwin Engine + Arena Brain
Darwin Engine v2 + Arena Brain (RAG) + Agentic RAG
Trois systemes revolutionnaires de recherche IA propulsent Strategy Arena. Inspires par l'autoresearch de Karpathy et la technologie RAG.
Voir l'IA Penser en Direct →
April 3, 2026 v7.1 โ€” Arena Autoresearch
Arena Autoresearch โ€” 2 000 Experiences de Strategies Chaque Nuit (Inspire par Andrej Karpathy)
Inspire par l'autoresearch d'Andrej Karpathy โ€” la meme boucle d'experimentation autonome qui a decouvert 20 optimisations LLM en 700 experiences, adaptee a l'evolution de strategies de trading. Notre systeme mute des strategies ArenaScript, les backteste sur BTC, garde les gagnantes. 2 000 experiences par nuit. Zero cout API.
April 3, 2026 v7.0 โ€” ArenaScript Launch
ArenaScript โ€” Le Langage de Trading Propre a Strategy Arena
Comme Pine Script pour TradingView, mais plus simple et plus puissant. Ecrivez des strategies en langage naturel avec Invictus, Chimera et detection de regime integres.
Essayer ArenaScript →
April 3, 2026 v6.5 โ€” Open Arena Protocol
Open Arena โ€” Amenez Votre IA Locale. Defiez 60 Strategies.
Connectez votre LLM local (Llama, Qwen, Mistral sur RTX 4090, Mac Mini M4) et tradez en live contre nos 60 strategies IA.
Rejoindre l'Open Arena →
April 3, 2026 v6.3
Mon IA + Bien Demarrer + Affilies Amazon
April 3, 2026 v6.1 โ€” Speed Optimization
Vitesse Mobile : 7.2s โ†’ 250ms (96% plus rapide)
Optimisation massive de la performance. Les pages s'affichent quasi instantanement sur mobile. Core Web Vitals Google ameliores sur tout le site.
April 3, 2026 v6.0 โ€” TradingView Integration
Convertisseur Pine Script + Analyse IA des Charts + Arena Upgrade
Importez n'importe quel Pine Script TradingView, convertissez-le avec Claude, et regardez-le combattre 60 strategies en live. Claude navigue aussi sur TradingView et lit vos charts.
Essayer le Convertisseur →
April 2, 2026 v5.5 โ€” US Market Launch
Version Anglaise Complete โ€” 20 Routes, 100 URLs, SEO Optimise
Strategy Arena est maintenant entierement disponible en anglais avec des routes /en/ dediees, 83 articles blog, et une optimisation SEO complete pour le marche US.
Visiter la page d'accueil →
April 1-2, 2026 v5.2
Academy Complete + Roast Battle + Wallet Oracle
Commencer l'Academy →
March 9, 2026 v5.0 โ€” Product Hunt Launch
๐Ÿš€ Claude LIVE โ€” Real-Time AI Trading
Claude Haiku now trades LIVE in the arena, calling the API in real-time for every decision. No pre-coded rules โ€” pure AI reasoning on every candle.
March 9, 2026
๐Ÿ“ฐ AI Insights โ€” Live Editorial Analysis
A new magazine-style page where Claude Haiku writes dynamic editorial summaries of what's happening across ALL arenas. Updated regularly, multilingual, and deeply educational.
March 9, 2026
๐ŸŽฌ YouTube Strategy Tester
Paste any YouTube trading video title & description โ€” Claude Haiku extracts the trading rules, auto-generates a strategy, and backtests it against Buy & Hold. 5 preset examples included.
March 9, 2026
Opus v6 โ€” New Claude Strategy
Claude Sentinel (v5) retired at -57% and replaced by Opus v6 โ€” a multi-confluence strategy inspired by the top 3 performers (CUDA Evolved, Smart Money, Deep Shadow).
March 9, 2026
ML Arena Reset & Stability Fixes
ML Arena models were experiencing capital overflow (LightGBM reached $5.6 billion). All 8 models reset to $10,000 with runtime caps to prevent future explosions.
March 7, 2026 v4.0
Community Arena
Submit your own trading strategy and compete live against other traders. Top performers get promoted to the main arena โ€” like a Ligue 2 to Ligue 1 system.
Join the Arena →
March 7, 2026 v3.5
Market Regime Performance
Each strategy page now shows how it performs in bull, bear, and sideways markets. Inspired by Reddit community feedback.
See Example →
March 7, 2026 v3.4
Multi-Strategy Portfolio Builder
Combine 2-8 strategies with custom weights. Backtest the combined portfolio and analyze diversification with correlation matrix.
Try Portfolio Builder →
March 7, 2026
Smart Money Concepts Strategy
New ICT-based strategy added to the arena: Order Blocks, Fair Value Gaps, Liquidity Sweeps, Break of Structure.
View Strategy →
March 7, 2026
New Blog Articles
Read Blog →
March 28, 2026 v5.0
🛡 INVICTUS โ€” The Arena's Immune System
Invictus learns from every losing trade across all 60 strategies. It feasts on defeats and grows stronger. 3 phases: Learn, Protect, Ascension.
See Invictus Live →
March 27, 2026 v4.5
🏛 Genie Pantheon โ€” 6 AI Oracles Debate Live
Ask any trading question. Claude, Grok, GPT, Gemini, DeepSeek, and Perplexity debate in parallel and give you a consensus verdict in 6 seconds.
Try Pantheon →
March 27, 2026 v4.4
🌊 Leviathan Awakening + Security Shield
March 27, 2026 v4.3
🦮 6 New Mathematician Pages + Verdict
Explore Minds →
March 26-27, 2026 v4.2
🤖 Local AI Strategies + Bot Intelligence
March 26, 2026 v4.1
⚡ Performance Fixes + Battle Royale Talk
March 24-25, 2026 v4.0
🌌 Landing Rebuilt + BTC vs Gold + God Mode
March 20, 2026 v3.5
🔮 Predictions + Polymarket + API + Ko-fi
March 7, 2026 v3.2
YouTube Strategy Tester
Paste any YouTube trading strategy description. Our AI extracts the rules and backtests them on real crypto data. Does it actually work?
Try YouTube Tester →
March 7, 2026 v3.3
Overfitting Detector
Every strategy gets a robustness grade (A-F). Is it truly robust or just overfit to historical data?
Try Overfitting Detector →
March 7, 2026 v3.1
Monte Carlo Simulation
Run 1,000 bootstrap simulations on any backtest to test strategy robustness. Does your strategy survive randomness?
Try Monte Carlo →
March 7, 2026
Multilingual Support & Technical Fixes
March 7, 2026
Reddit Content Intelligence
March 6, 2026 v3.0
Chimera Brain V2
The Chimera meta-strategy got a massive brain upgrade with GPU-computed pattern matrices.
February 2026 v2.8
Strategy Genie
Ask our AI anything about trading strategies. Get personalized recommendations based on your risk profile.
Try Strategy Genie →
February 2026 v2.7
60 Strategy Detail Pages
February 2026 v2.6
5-Language Support
February 2026 v2.5
Backtesting Pro + Strategy Builder
Full backtesting engine with configurable strategies, fees model, and no-code strategy builder.
Try Backtesting →
February 2026 v2.4
Security Audit & Infrastructure
January 2026 v2.0
The Arena Goes Live
60 strategies compete in real-time across multiple crypto assets. The arena never sleeps.
See the Arena →

Comment fonctionne le cerveau IA de Strategy Arena

La plupart des plateformes de trading IA font des appels API isoles avec des prompts statiques. Chaque appel est independant, sans memoire et sans conscience de ce que les autres composants ont decouvert. Strategy Arena fonctionne autrement.

PromptForge est le moteur de contexte central. Avant chaque appel IA (14 modules), PromptForge injecte 12 sources de donnees en temps reel : regime de marche, RSI, top strategies, donnees de survie Invictus, patterns Chimera (1 221 patterns), recommandations ML Hydra, signal Leviathan, sentiment news, lecons Wiki issues de la recherche nocturne, hypotheses strategiques, decouvertes Hall of Fame, et statut du filtre Nutrition.

Component Memory donne a chaque module une memoire persistante. Quand l'Oracle vote sur une question, il se souvient de ses consultations precedentes. Quand l'arene Collaborative delibere, chaque IA connait ses votes anterieurs et si le consensus etait correct ou non. Cette memoire s'accumule et est lisible par Hermes, l'agent LLM local qui effectue des analyses inter-composants.

La boucle Karpathy ferme le cercle. Chaque nuit, 11 moteurs autonomes executent des milliers d'experiences : Darwin evolue les parametres, Meta-Harness optimise Darwin lui-meme, Chimera evolue ses seuils de patterns, et le dernier moteur evolue le prompt de chaque IA selon son taux de precision. Les resultats alimentent le Living Wiki, qui re-alimente PromptForge, creant un systeme d'intelligence en boucle fermee qui s'ameliore pendant que vous dormez.

Cette architecture est inspiree par ActiveWiki, notre framework open-source pour les systemes de connaissance en boucle fermee, et par le pattern autoresearch d'Andrej Karpathy : accumuler → hypothese → tester → apprendre → repeter.

Rejoindre le canal โœ• ๐Ÿ’ฌ Feedback