Live benchmark on real data

ActiveWiki Benchmark

We built a framework. Then we tested it on our own platform. Here are the results — no cherry-picking, no marketing, just data.
Tested on 2,530+ real experiments
Hypotheses
vs 6 existing
Consolidated
vs N/A
Cycles
Lessons
Graph Nodes
Feature Existing System ActiveWiki
Hypotheses per run619 (3.2x)
Detection strategies5 (Python rules)12 (including temporal, counterfactual, meta)
Counterfactuals✅ Challenges strongest beliefs
Knowledge Crystallization✅ 3+ lessons → meta-knowledge
Self-Reflection✅ Auto-tunes decay_rate + max_hypotheses
Confidence ScoringLabel only (high/medium/low)0-1 numeric, evolves with time
Expected Impact✅ ROI-like score per hypothesis
HTML Dashboard✅ Auto-generated every cycle
Research Brief✅ Auto-published every 7 cycles
Wiki Pruning✅ Intelligent page cleanup
Hypothesis Evolution✅ Evolves old hypotheses into v2.0
Cost$0$0
DependenciesCustom PythonZero (pip install activewiki)

Auto-Generated Research Brief

ActiveWiki auto-publishes a mini research paper every 7 cycles. Here's the latest:

Loading research brief...

How We Tested

ActiveWiki was installed on the same VPS that runs Strategy Arena. It ingested the same data (Darwin Engine results, Living Wiki lessons, nightly logs from 2,530+ experiments). It ran 3 cycles with a simulated backtest engine. No data was modified. The existing system continued running normally.

The comparison is fair: same data, same machine, same moment. The only difference is the framework processing it.

View on GitHub Memory Stack Evolution Lab

ActiveWiki is an open-source framework (MIT license) that implements a closed-loop scientific knowledge system. Tested in production on Strategy Arena's 2,530+ nightly experiments, it generates 3.2x more hypotheses than the existing Python-based system while adding counterfactual simulation, knowledge crystallization, self-reflection with auto-tuning, confidence scoring, and auto-generated research briefs. Built by the Strategy Arena team where 59 AI trading strategies evolve autonomously every night. The framework is domain-agnostic: applicable to trading, code quality, SEO, research, and any field where knowledge should generate action. GitHub repository.

Rejoindre le canal 💬 Feedback ☕ Support Us