The Market Microstructure team studies the mechanics of how prices form, how liquidity flows and how different market participants leave footprints in the data. This is not technical analysis in the traditional sense; it is the study of market physics.

Regime detection is a cornerstone of this work. In Experiment 1, 990 strategies using standard pivot points all failed when volatility collapsed by a factor of three. They were optimized for a world that no longer existed.

Volume analysis and pivot structures reveal invisible walls, price levels where VPOC, Fibonacci, CPR and multi-timeframe pivots create confluence. These walls are invisible to standard pivot analysis but create measurable resistance and support.

The key finding from Experiment 2: a strong regime gate combined with high confluence produces the best validation survival. Of 888 strategies tested with regime awareness and money zone detection, 12 survived rigorous out-of-sample validation.

The Learning Systems team is developing metacognition, the ability for machines to monitor, evaluate and improve their own learning process.

The framework defines five capabilities: NOTICE, QUESTION, PRIORITIZE, EXPERIMENT and JUDGE. These allow the system to detect prediction errors, form hypotheses, allocate attention, test ideas and update its learning model.

A critical distinction is Knowledge versus Experience. Knowledge is external, acquired and unvalidated. Experience is PE-validated, earned through the machine’s own prediction and error cycle.

Regime-Indexed Recall makes experience permanent while activation depends on regime similarity, not temporal decay. A lesson from a high-volatility regime three years ago becomes instantly accessible when a similar regime returns.

The Causal Inference team applies the structural causal models framework to ensure systems reason about cause and effect, not merely co-occurrence.

The causal ladder has three rungs: Association, Intervention and Counterfactual. Standard machine learning mostly operates at association. Autonomous decision systems require intervention and counterfactual reasoning.

The Causal Ledger is a formal record of every causal claim, its evidence basis and its confidence level. When the system believes A causes B, the ledger records why, what evidence supports it, what could refute it and when the claim should be revisited.

The Hub-and-Spoke architecture serves causal inference directly. Independent sensory spokes feed a convergence hub that detects agreement, conflict and novelty. When spokes conflict, that conflict is the important information.

The Safety and Alignment team ensures that autonomous systems behave as intended, remain auditable and fail safely. This is not a compliance exercise; it is a core engineering discipline.

The Contrarian Gate is the first line of defense. Every proposed action must survive adversarial questioning: what is the obvious approach, why might it be wrong and what would a sophisticated adversary exploit?

The Reducibility Detector estimates whether a phenomenon is tractable before the system invests resources. Some market phenomena may be computationally irreducible and cannot be predicted without simulation.

The Cryptographic Pre-Registration protocol hashes and timestamps strategy parameters, predictions and success criteria before execution, preventing retroactive overfitting claims.