Reading the Crypto Pulse
How we predict daily crypto direction on Polymarket by combining multi-signal analysis with disciplined position sizing and correlation intelligence.
Every day, Polymarket runs prediction markets asking a simple question: "Will Bitcoin (or Ethereum) go up or down today?" Traders buy and sell shares on the outcome, and the share price reflects the crowd's collective belief. Our job is to figure out when the crowd is wrong.
It is like reading barometric data before a sailing race. The crowd sees blue skies and bets accordingly. But if you have wind pattern models, satellite imagery, and pressure readings, you might spot the storm front they are missing. We do not try to predict crypto prices. We build calibrated probability estimates from dozens of market signals and act when our estimate diverges meaningfully from what the market is pricing.
This page explains how our BTC and ETH models think, from raw signals to the picks that surface in your terminal.
The Challenge
Why Crypto Direction Is Uniquely Hard
Imagine predicting whether a river will rise or fall tomorrow. You could look at the water level right now, but that tells you almost nothing about tomorrow. The answer depends on rainfall upstream, snowmelt in the mountains, dam releases, and a dozen other factors — many of them invisible from the riverbank. Crypto markets work the same way: the current price is just the surface.
24/7 Volatility
Crypto never closes. A single overnight headline — a regulatory action, an exchange hack, a whale liquidation — can reverse the entire day's trend.
Regime Shifts
What drives crypto changes month to month. Sometimes it tracks the stock market tick-for-tick. Other times it decouples entirely. A model must know which regime it is in.
Efficient Market
Polymarket attracts sharp crypto-native traders. The obvious patterns are already priced in. Finding edge requires looking where others are not.
The opportunity: Most Polymarket traders rely on surface-level reads, Twitter sentiment, or a single indicator. When our model synthesizes dozens of orthogonal signals — price structure, derivatives positioning, macro flows, on-chain behaviour, and regime awareness — we spot the moments when the crowd has mispriced direction.
The BTC Model
Multi-Signal Intelligence
Imagine you are a doctor diagnosing a patient. You would not prescribe treatment based on a single blood test. You would run a full panel: heart rate, blood pressure, temperature, lab work, imaging. Each test tells you something the others cannot, and the combination reveals the true picture. Our BTC model works the same way.
The model ingests data across multiple distinct signal categories every morning, engineers dozens of features from raw data, and feeds them into a trained machine learning classifier that has learned which combinations of signals predict daily direction.
| Signal Category | What It Captures |
|---|---|
| Price Structure | Multi-timeframe analysis of price behavior, trend dynamics, and volatility patterns. |
| Market Positioning | Signals derived from how traders are positioned in the derivatives market. Crowded trades create opportunities. |
| Capital Flows | Tracking where institutional and smart money is moving before the crowd catches on. |
| Macro Environment | Cross-market signals from equities, currencies, and broader risk sentiment. Bitcoin does not trade in a vacuum. |
| Network Intelligence | Proprietary signals derived from blockchain activity and network behavior patterns. |
| Regime Awareness | Automatically classifies the current market environment so the model adapts its approach accordingly. |
Key insight: Any single signal category produces mediocre predictions on its own. The edge comes from the combination: derivatives positioning confirming a price structure signal, supported by institutional flow and a favourable macro backdrop. The model learns these multi-dimensional interactions from historical data.
Calibrated Probabilities
The model does not just predict UP or DOWN. It outputs a calibrated probability: the likelihood that BTC will finish up by noon ET tomorrow. This probability drives every downstream decision: whether to play, how much to risk, and what edge we have over the market.
Calibration means when the model says 60%, the outcome is UP roughly 60% of the time. Overconfident models blow up. Calibrated models compound.
Regime-Aware Weighting
Markets shift between distinct regimes — macro-driven, crypto-native, trending, and mean-reverting. The model detects the current environment and adjusts its approach accordingly.
A signal that works in trending markets may fail in mean-reverting ones. Regime awareness prevents the model from using yesterday's playbook in today's environment.
The ETH Model
Correlation Intelligence
BTC and ETH are two ships in the same harbour. Most days, they rise and fall with the same tide. But occasionally, ETH breaks away — a protocol upgrade, a DeFi event, or a shift in altcoin sentiment sends it on its own path. The question is: when can we trust the tide, and when should we sit out?
Our ETH model answers this by monitoring the real-time correlation between BTC and ETH. When the BTC model produces a high-conviction signal and the two assets are moving in lockstep, we extend the signal to ETH with adjusted position sizing. When correlation breaks down, we stay on the sidelines.
| Filter | What It Does |
|---|---|
| BTC Signal Strength | Only relays when the BTC model has high-conviction directional signal. Weak signals are not worth extending. |
| Correlation Filter | Monitors the real-time relationship between BTC and ETH across multiple time horizons. The filter must confirm alignment before any relay. |
| Confidence Mapping | Maps the BTC model's probability to a relay confidence tier that determines display and tracking. |
| Conservative Sizing | ETH positions are sized conservatively relative to BTC. The correlation add-on should always be the smaller companion bet. |
Why this works: BTC and ETH are strongly correlated most of the time. By combining the BTC model's validated edge with a correlation filter that sits out during decoupling events, the ETH model achieves positive expected value while dramatically reducing the risk of catching an ETH-specific move the wrong way.
Sit-Out Discipline
The most important feature of the ETH model is knowing when not to play. When correlation drops below the minimum threshold, the model produces no pick. This is deliberate: the times when ETH and BTC diverge are precisely the times when extending the BTC signal would be most dangerous.
In backtesting, the sit-out filter improved accuracy by removing the noisiest days from the portfolio. Fewer picks, higher quality. The same principle applies across all our models: discipline in selection is more valuable than volume.
Position Sizing & Risk
Kelly-Criterion Sizing ($100 Units)
Imagine a card counter at a blackjack table. They do not bet the same amount every hand. When the count is high (strong edge), they bet more. When it is low, they bet the minimum. Our crypto models use the same principle: position sizes scale with edge strength, governed by the Kelly Criterion.
| Metric | How It Works |
|---|---|
| Unit Size | $100 per unit. Each unit represents a fixed dollar amount for consistent tracking and bankroll management. |
| BTC Sizing | Position size scales with edge strength — higher conviction means more units, governed by a mathematically optimal framework. |
| ETH Sizing | Conservatively sized relative to BTC. The correlation relay should always be the smaller companion bet. |
| Hard Cap | Maximum position size enforced regardless of model confidence. Prevents overexposure on any single day. |
Expected Value (EV): Displayed as a dollar amount on a $100 stake. If a pick shows +$12.80 EV, the model estimates you gain $12.80 in expected value for every $100 wagered. EV is the long-term compass: individual picks win or lose, but positive EV compounds over many bets.
The Confidence System
Multi-Gate Quality Control
Not every day produces a pick. Like a telescope, you can only observe the stars on a clear night. On cloudy nights, you pack up and go home. Our crypto models apply multiple independent filters before surfacing a pick:
Gate 1: Minimum Probability
The model's calibrated probability must exceed a minimum threshold. Below this, the signal is too noisy to act on.
Gate 2: Edge vs. Market
The model's probability must meaningfully exceed what the market is pricing. Marginal edges get rejected.
Gate 3: Regime Check
Certain regime conditions automatically reduce confidence or block picks entirely. The model does not play in environments it has not validated.
Gate 4: Correlation (ETH)
For ETH picks, both short-term and medium-term BTC/ETH correlation must exceed minimum thresholds. No correlation, no pick.
How We Measure Success
Crypto is volatile. Individual days will lose. We measure the system, not the day.
Directional Accuracy
Percentage of picks where the model correctly predicted the day's direction. We track overall accuracy and accuracy by confidence tier separately.
ROI on $100 Units
Return on investment across all picks with $100 unit sizing. This accounts for position sizing, not just win rate.
Calibration
When the model says 60% UP, does it hit 60%? Calibration drift is the first sign a model is degrading and triggers a review.
Streak Tracking
Current win/loss streak and historical streak distribution. Helps distinguish normal variance from model breakdown.
The Bottom Line
Crypto direction prediction is one of the hardest problems in quantitative finance. The edge is thin, the noise is enormous, and the market is smart. Our approach is built on a simple philosophy:
- Synthesize, do not simplify: dozens of orthogonal signals across price, derivatives, flows, macro, and on-chain data — not one indicator
- Calibrate ruthlessly: probabilities must reflect reality, not model optimism. Overconfidence is the silent killer
- Sit out when uncertain: no pick is better than a bad pick. The models produce no signal on many days, and that is by design
- Size for survival: Kelly-criterion sizing with hard caps. The goal is to compound over months, not to win today
- Track everything: full transparency on accuracy, ROI, calibration, and streaks so you can evaluate us with real data
The market is noise. Our job is to find the signal within it.
Technical Breakdown
For the quantitative readers: a deeper look at the architecture and workflow.
Machine Learning Classifier
The BTC model uses a proprietary machine learning classifier trained on engineered features across all signal categories. Walk-forward validation ensures the model never sees future data during training.
Predictions are calibrated post-training to ensure probability outputs reflect observed frequencies.
ETH Correlation Engine
The ETH model monitors the rolling relationship between BTC and ETH across multiple time horizons. Both must confirm alignment before a pick is generated. Position sizing is conservative relative to the primary BTC signal.
The correlation methodology is designed to avoid spurious signals and only acts when the relationship is genuinely strong.
Model Workflow
6:00 AM ET: BTC pipeline runs — data collection, feature engineering, model prediction, pick publishing
6:15 AM ET: ETH relay evaluates — reads BTC prediction, checks correlation, publishes relay pick if filters pass
7:00 AM ET: Picks lock and surface in your terminal
12:10 PM ET: Resolution — Polymarket markets settle, picks are graded, performance updated
Monthly Access
- Predictions only go live when the model finds true edge
- Closing line value tracked on every prediction so you can verify it yourself
- Covers every market we model and we're always adding more
- Cheaper than your average unit size
Annual Access
- Get 4 months free on us when you go annual
- Every new model we ship is included automatically
- Full platform access for less than most services charge monthly
- Models run 365 days, your subscription should too