Hybrid Forecasting Intelligence | Dysrupt Labs - Reliable departure from human and AI consensus

Consensus Divergence you can Trust

Problem: In a world drowning in consensus, edge lives in divergence. AI models replicate it instantly. Sell-side research amplifies it endlessly. The scarce commodity is validated divergence — the trustworthy signal that the crowd is wrong.
Solution: Dysrupt Labs reliably detects divergence from consensus. In US macro it maps them into systematic signals for US FX pairs and ETFs.
Method: Combine human judgment at scale with machine learning that identifies real-time informational edge. Operating since 2008 across geopolitics, elections, and disease, we've spent three years analysing 35,000 hours of continuous market microstructure and mapping this divergence system into US FX pairs and ETFs where it delivers CIO-grade performance.
Result: A rigorously validated feed of early-warning divergence signals. Not more noise. Not more consensus. Just disciplined divergence you can use.
Dysrupt Labs: is a team of 12 researchers passionate about forecasting. Our hybrid human-machine forecasting system solves the fundamental problem: distinguishing between who's historically accurate versus who's informed right now. Serving family offices and institutions with validated early-warning signals and alternative data not available through traditional channels. Reliable departures from human and AI consensus.
Almanis: is our platform forecasting geopolitical, election and disease events. Our recent epidemic forecasting research has evolved into a systematic approach for identifying when market consensus is likely to be wrong on key macroeconomic variables. This is driven by a rigorously curated global forecasting network of 900+ contributors in a highly automated system operating at scale with Four 9s efficiency since 2015. We identify the human judgment that pattern-matching AI lacks, while our 43-feature ML system identifies which forecasters have genuine insight in real-time—not just historical track records.
Serving: as a private intelligence firm for family offices and institutions, our business operates through direct relationships and a small digital footprint, ensuring confidentiality for both our forecaster network and institutional partners. We provide what sophisticated investors need most: validated signals of when everyone else is wrong. We maintain discipline—signaling only when our analysis diverges significantly from future expectations of humans or AI.
900+
Global Forecasters
17
Years Operating
99.99%
System Uptime
0.8/day
Signal Frequency (with 20 to 100 assets per signal)

Human + Machine > AI + AI

The 6-Step Process AI Cannot Replicate

"AI can replicate consensus almost instantly. It cannot yet generate the judgment, narrative, and conviction required to build edge... The scarce resource is no longer content. It is trust." - Jason DeRise, The Data Score
1

900+ Human Forecasters Make Judgments AI Cannot

Our expert network provides real expertise, not pattern matching. They assess unprecedented situations that no training data can capture—the kind of qualitative judgments that create genuine market insight.

2

43 Behavioral Features Analyzed Per Forecast

Our ML systems analyze 43 distinct behavioral patterns in real-time trading activity, identifying the difference between the informed and noise. This micro-level analysis reveals which human judgments contain genuine information.

3

ML Identifies Which Humans Have Genuine Insight

We don't track who's been historically accurate—we identify who's informed now. Our algorithms detect which forecasters have real-time information advantage versus those following patterns or consensus.

4

92% Agreement = No Signal

When our hybrid system agrees with consensus, we generate no signal. We ignore consensus confirmation entirely. This discipline separates us from AI systems that generate infinite content regardless of value.

5

8% Divergence = Trade Signal

Only when we diverge significantly from consensus (>1.65σ) do we issue tradeable alerts. This selectivity—signaling ~0.8 times per day versus AI's infinite daily reports—is where our edge lives.

6

72.7% Accuracy on Divergence

We achieve 72.7% accuracy (which means when we disagree with consensus, we are right 7 in 10 times) precisely when consensus is wrong. This is the only metric that matters in a market saturated with identical analysis—being right when everyone else is wrong.

Market Application

After three years and 35,000 hours of live backtest data, Dysrupt Labs maps validated divergence into tradeable signals across US FX pairs and US ETFs.

≥65%
Hit Rate
≥1.6
Sharpe Ratio
≤5%
Max Drawdown
<500ms
Latency
≥25%
Error Reduction

CIO-Grade KPI Benchmarks

Metric Benchmark Note
Hit Rate ≥ 65% On divergence-classified events
Sharpe Ratio ≥ 1.6 Scales > 2.5 at higher thresholds
Max Drawdown ≤ 5% Risk discipline maintained
Latency < 500ms Feed-to-alert
Error Reduction vs Consensus ≥ 25% Systematic uplift
Geopolitics Elections Disease Macro Data Prints US FX Pairs US ETFs

This is not "alt-data." It is a validated signal engine — institution-ready.

Independent Validation

  • DARPA NGS2 Program: Methodology successfully replicated using out-of-sample datasets
  • eBioMedicine (The Lancet): ML-augmentation framework peer-reviewed and published
  • Journal of Financial Markets: Core forecasting methodology published
  • Operational Pedigree: 17 years across geopolitics, elections, and disease forecasting with Four 9s uptime
  • Family Office & Institutional Trials: Previous trials with major hedge funds and family offices informed our systematic approach to signal generation
  • 15-Year Track Record: Operational history across multiple forecasting domains
  • Curated Global Network: 900+ contributors maintained through continuous performance evaluation and relationship management across 15+ countries

Leadership

Karl Mattingly, Founder & CEO

25-year ANZ Bank career in senior risk and international banking, followed by 17 years developing collective intelligence systems. MBA Columbia University.

Prof. Anne-Louise Ponsonby, Founder & Chief Scientist

Principal Investigator at University of Melbourne and The Florey Institute. Leading scientific validation and academic collaboration efforts. In addition to Dysrupt, Anne-Louise is a foundation investor in a number of successful biotech startups. Her expertise includes medicine, epidemiology, quantitative methods, and "dirty" data.

Stephen Markscheid, Founder & Director

35+ years international experience at Boston Consulting Group and General Electric. MBA Columbia, MA Johns Hopkins SAIS. In addition to Dysrupt, Stephen is an active private investor and director. Expertise in corporate advisory.

Ian Clark, Founder, CFO & COO

Chartered accountant with honours in Economics from Australian National University. Past roles include PwC Partner, CFO of Australian listed technology company and Federal Government Agency. In addition to Dysrupt, Ian is an active private investor. Expertise in quantitative methods and real options.

Frequently Asked Questions

Is this just another AI consensus product?
No. Our system is designed to detect departures from consensus and alert only when divergence is statistically significant.
How noisy is the signal?
Approximately 0.8 alerts per day on average (with 20 to 100 assets per signal) — selective by design. We maintain discipline, signaling only when divergence exceeds 1.65σ.
Where can the signals be applied?
We currently map to US FX pairs and US ETFs; other markets can be evaluated under the partner framework.
Can we validate performance?
Yes. NDA → Data Room → internal replication. KPIs and MCA safeguards align capital deployment with verified results.
Why exclusive partnerships?
Divergence is valuable only when scarce. We preserve signal integrity through exclusivity, ensuring competitive advantage.
What happens when your divergence signal is wrong?
Our 72.7% accuracy means we're wrong approximately 3 in 10 times when we diverge from consensus. We provide full transparency on signal performance, including false positive analysis and drawdown periods. Each signal includes confidence intervals and historical performance metrics for similar divergence patterns. This transparency allows institutions to size positions appropriately and integrate our signals into broader risk management frameworks.
How do you prevent signal decay?
Signal decay is inherent in consensus-following strategies, but our approach has three structural defenses. First, we maintain exclusivity through limited partnerships rather than broad distribution—preserving scarcity by design. Second, our signals identify divergence from future consensus, not current patterns. As markets evolve and new consensus forms, our human forecasters adapt their judgment in real-time, while our ML continuously recalibrates which forecasters have genuine insight versus those now following the herd. Third, we only signal on high-conviction divergence (>1.65σ), approximately 0.8 times daily. This selectivity means we're not mining marginal edges that decay quickly, but identifying fundamental disagreements with consensus. The edge isn't in the signal itself—it's in knowing which humans are genuinely informed right now versus those pattern-matching yesterday's winners.
How do signals integrate with existing investment processes?
Our signals are designed for seamless institutional integration through three delivery mechanisms. First, real-time API feeds compatible with major OMS/EMS platforms, delivering structured JSON with asset identifiers, divergence magnitude, confidence intervals, and historical performance context. Second, customizable alerting via webhook, email, or messaging platforms, with filtering by divergence threshold, asset class, or confidence level. Third, a dashboard interface for signal analysis, performance attribution, and back-testing capabilities. Signals include standardized risk metrics allowing direct integration with position sizing algorithms and portfolio optimization systems. We provide a 30-day parallel run period where institutions can shadow-trade signals alongside existing strategies, validating performance and refining integration parameters. Our technical team provides white-glove onboarding support, including custom API endpoints, historical signal data for back-testing, and integration with proprietary risk management frameworks. The implementation typically takes 2-3 weeks from NDA to live signals, with most institutions running signals at 10-30% of target allocation during initial validation before scaling to full deployment.

Client Advantage

We provide sophisticated family offices and institutions with validated signals when consensus is systematically wrong. Our approach to partnerships is built on transparency, rigorous methodology, and the recognition that in a world drowning in AI-generated consensus, the life raft is divergence detection that you can trust.

Contact: karlmattingly@dysruptlabs.com or kellybanks@dysruptlabs.com