Newcomer Guide

Dysrupt Labs is privately held, founder-employee owned, based in Melbourne Australia, established in 2008. The company operates with approximately 12 FTE across research, platform engineering, and forecaster operations. The core asset is Almanis, a private prediction market with 900+ expert forecasters carefully curated from over 36,000 candidates across 17 years of operation. The median forecaster tenure is 7+ years, and the environment is KYC-managed for participant quality control.

What is a Prediction Market?

A prediction market is a financial exchange where participants can buy and sell contracts whose payout depends on the outcome of a future event. These contracts aggregate distributed information from thousands of participants into a real-time probability estimate. The mechanism is identical across platforms: individual beliefs are aggregated, price discovery happens through competition, and the final price represents the collective forecast.

As of 2025, the combined notional volume on public prediction markets (Kalshi and Polymarket) exceeded $44B. The sector is expanding rapidly: ICE Futures launched Polymarket Signals in February 2026, and projections suggest the market will reach $222B in 2026. Dysrupt Labs operates the largest private prediction market in the world, serving as a research platform and real-time signal pipeline.

Public vs. Private Markets

What Happens When a Forecaster Trades

Dysrupt Labs uses the Logarithmic Market Scoring Rule (LMSR) mechanism, an automated market maker that incentivises honest probability revelation. When a forecaster trades, they are updating the collective forecast. The mechanism ensures that each participant faces the true marginal cost of moving the price—they cannot profit by submitting false beliefs.

The LMSR is mathematically elegant: it preserves subsidy predictability (the host's maximum loss is known in advance), allows continuous price updates, and generates natural hedging opportunities. It is documented in peer-reviewed research by Robin Hanson and has been independently validated across multiple institutions including DARPA-funded programmes and commercial platforms.

Why Macro?

Dysrupt Labs focuses on macroeconomic events—CPI, GDP, NFP, PCE, Retail Sales, and Housing Price Index. This focus is deliberate, based on three core criteria:

Recurring: Macro data releases follow regular schedules. CPI is monthly, NFP is monthly, GDP is quarterly. Recurring events allow for systematic backtesting and signal validation across dozens of episodes.

Unambiguous: The outcome is binary and verified by official sources (BLS, Bureau of Economic Analysis, Federal Reserve). There is no dispute about ground truth, eliminating resolution risk.

Tight Asset Linkage: Macro events move FX and commodity-linked ETFs predictably. This creates the execution bridge from the forecast signal to tradeable positions.

Note: Dysrupt Labs operates no markets involving listed equities (MNPI risk) or events involving assassinations or death counts (ethical values decision).

The prediction market produces three concurrent signals, each capturing different layers of information from the forecaster network.

Signal 1 is the public forecast. Every prediction market in the world produces it. Signals 2 and 3 require private access to the forecaster microstructure—only available on Almanis.

Most of the time, Signals 1 and 2 agree. This is "Confirmation"—95% of all trading days. The crowd consensus and the insight signal point in the same direction. Nothing unusual is happening.

Periodically, they diverge. Approximately 5% of trading episodes, Signal 2 materially separates from Signal 1. The insight cohort is pricing the outcome differently than the aggregate. This is when the diagnostic becomes active.

Practical Example: CPI Forecast

On day 1, all 900 forecasters are converging on a CPI print of 3.4%. Signal 1 is 3.4%. Signal 2 is 3.4%. The z-score is 0. Confirmation.

On day 5, as new economic data accumulates, 880 forecasters shift to 3.3%, but 20 forecasters (including several with strong track records in high-uncertainty regimes) remain at 3.5% and are actively trading against the shift. Signal 1 is now 3.3%. Signal 2 is 3.5%. The z-score spikes to +2.1σ. Divergence alert fires. A trade is executed long the USD (betting on the higher CPI outcome that the insight cohort is pricing).

On day 7, the BLS CPI release comes in at 3.5%. Signal 2 (and the cohort) was correct. The trade profited. Divergence resolved.

The distinction between Risk and Knightian Uncertainty, formulated by economist Frank Knight in 1921, is fundamental to understanding where prediction markets work and where they reach their boundary.

Risk is calculable. It is the domain of known probabilities, measurable volatility, and historical frequency. You can estimate risk using historical data, volatility models, and distributional learning. This is where AI excels. A neural network trained on 50 years of CPI data, employment dynamics, and consumer behaviour can generate accurate probability distributions for next month's CPI.

Knightian Uncertainty is non-calculable. It is the domain of novel events, regime shifts, regime-conditional asymmetries, and phenomena whose probability structure itself may be changing. You cannot estimate Knightian uncertainty by scaling data or adding layers to your model, because the underlying distribution is not stationary. A pandemic is not a tail event in a normal distribution—it is regime change. A financial crisis is not volatility clustering—it is a fundamental shift in how markets relate to fundamentals.

The Rumsfeld Matrix—known knowns, known unknowns, unknown knowns, unknown unknowns—is Dysrupt Labs' own framework, not Knight's original formulation. It maps naturally onto the prediction market signal cascade:

Sentinels vs. Knights

Inside stable regimes (Risk domain), the crowd consensus (Signal 1) is highly accurate. We call these regime phases "Sentinel" periods. The crowd is seeing clearly. AI sees the same thing faster and cheaper.

At the boundary (Knightian Uncertainty domain), the consensus breaks down. The regime itself is shifting. Distributional learning fails because there is no stable distribution to learn. This is when the small cohort of forecasters with superior pattern recognition at regime boundaries produces disproportionate accuracy. We call them "Red Knights"—a term that disrupts the medieval imagery (hierarchy, feudal obligation) that the bare word "Knight" retrieves, and activates three functional traditions: De Bono's Red Hat (intuitive, pre-articulate judgment), Red Team (institutionalised adversarial dissent), and Frank Knight (the uncertainty boundary).

The diagnostic operates at the transition from Sentinel to Knight—it detects when the regime is shifting and which forecasters are correctly pricing that shift.

The forward test translates signals into executed trades. The execution protocol is mechanically disciplined, with no discretion or override allowed.

Entry Criteria

A trade enters when Signal 3 (the divergent z-score) spikes to ±1.65σ or higher. This threshold is statistically significant and historically reliable—it corresponds to the moment when the insight cohort's divergence from consensus is large enough and stable enough to warrant execution.

The entry window is 5 minutes from the alert. The trader must execute within this window or pass. No queuing, no waiting for better fills.

Exit Criteria

A trade exits under three conditions, whichever comes first:

1. Z-Score Reversion: When Signal 3 reverts to ±0.5σ, the divergence has contracted and the insight advantage has disappeared. Exit immediately.

2. Macro Release Minus 60 Minutes: When the scheduled macro event is 60 minutes away, exit all positions. The event outcome is imminent and the signal is about to be resolved. No holding through the event.

3. Stop Loss: A 5% stop-loss is maintained on all positions as a circuit breaker. If the position moves 5% against the signal, exit.

Execution Universe

FX Pairs: EUR/USD, GBP/USD, USD/CAD, USD/AUD, USD/NZD, USD/CHF, USD/MXN. Six majors plus one emerging-market pair for interest-rate sensitivity.

ETFs: BIL (short-duration US treasuries, 1-3 month duration), SPY (S&P 500 equity), DXY (US dollar index). These provide leverage and cross-asset diversification for the macro signal.

A controlled forward test commenced in March 2026 to validate the signals out of sample in real time. The test is structured in three phases over a 12-month horizon.

Phase Schedule

Phase 3A (Months 1–3): Paper Trading
31 March – 30 June 2026. All signals are generated, all trades are executed at live market prices, but no actual capital is committed. This period establishes the baseline performance without capital risk. Weekly reporting published via Substack with full transparency on signal hits, misses, and methodology.

Phase 3B (Months 4–9): Live Trading
1 July – 30 September 2026. Contingent on Phase 3A performance review and board approval. Real capital is deployed against live markets. Positions are executed, P&L is tracked, and results are compared to paper trading performance. Weekly reporting continues.

Phase 3C (Months 9+): Packaging for Investor Due Diligence
October 2026 onward. Results are compiled, analysed, and packaged for institutional investors and qualified strategic partners. The data becomes the foundation for deployment and licensing discussions.

Three-Pod Structure

The forward test operates three independent sub-accounts on Interactive Brokers, each with $1M notional. The three pods are completely separate—different entry decisions, different exit triggers, different performance tracking. This allows three real-world variations on the execution protocol to be tested simultaneously.

Pod separation eliminates model overfitting risk and ensures that results are robust to implementation variations.

Gate Assessment Framework

Progression through the phases is governed by strict gate assessment criteria. Phase transitions are data-driven, not discretionary.

Independent Verification

An independent third-party verifier conducts weekly reconciliation of all trades, P&L reporting, and methodology compliance. Monthly statements are issued to qualified evaluators. This ensures that forward test results are independently auditable and not subject to internal manipulation or selective reporting.

Protocol Discipline

The forward test observes strict protocol discipline:

– No deviations from the entry/exit rules. Once the z-score criterion is met, entry is mechanical.

– 5-minute order window enforced. No discretionary delays or repositioning.

– No trailing stops or discretionary exit modifications. Exit happens at the three specified conditions only.

– All trades are logged in real time with timestamps, entry prices, and exit prices. Nothing is reconstructed post-hoc.

Polymarket Signal Pipeline

As of March 2026, Signal 2 (divergence) and Signal 3 (scored divergence) have been successfully replicated on Polymarket's public microstructure. The Polymarket implementation uses a constant-product automated market maker (AMM), not the LMSR that powers Almanis.

This replication is significant: if it generates profitable trading results, it establishes that the signals are a property of human behaviour under Knightian uncertainty, not an artefact of any specific cost function (LMSR) or platform (Almanis). The working paper's scope condition on LMSR-specificity (Section 6.3) would be empirically superseded.

Preliminary indications suggest the Polymarket pipeline is tracking the Almanis signals with high fidelity. Formal confirmation pending Phase 3B execution results.

The three-signal architecture is grounded in four major publications, each representing independent validation or extension of the methodology.

Reference glossary of core terms used throughout the Dysrupt Labs signal architecture and forward test.