Our Journey
FX Regime Lab is not a product. It is an experiment in transparent macro research. Here is how we got here, and where we are going.
The Spark
Built the first Python script to scrape CFTC COT data and compute net positioning percentiles. A single Jupyter notebook. No frontend. No database.
First Composite
Combined rate differentials and COT into a weighted composite score. Manual Excel tracking. First realization that static weights felt arbitrary.
The Pipeline
Built the first automated pipeline with Prefect. Added vol, OI, and cross-asset signals. Supabase backend. Still no public frontend.
Web Terminal
Launched the first web interface. Added track record visualization, daily briefs, and desk open cards. Still learning that our accuracy was near-random.
Three-Layer Engine
Introduced Layer 1 (structural gate), Layer 2 (directional conviction), and Layer 3 (execution HUD). Added hysteresis, Marcus clash logic, and confidence scoring. First comprehensive methodology document drafted.
Radical Honesty
Before public launch, we conducted a deep mathematical audit. Discovered that our confidence scores were not probabilities, our RR proxy was synthetic, and our accuracy was statistically indistinguishable from random. Rather than hide this, we published our limitations openly.
Proper Probabilities
Isotonic regression calibration for confidence scores. Real 25Δ risk reversal data. Bayesian dynamic betas via Kalman filter. Walk-forward ICIR weight optimization. Hidden Markov Model for regime detection. This is when the signal architecture becomes statistically defensible.
Expand the Universe
Add GBP/USD, AUD/USD, USD/CAD, USD/CHF as active pairs. Each pair will require its own calibrated model and weight optimization.
The Open Standard
Publish the full methodology as an open standard. Invite external auditors. Host a public model competition. Prove that transparent macro research can be conducted at institutional standards without institutional budgets.