Help Center
Product guide, feature reference, and key concepts.
Published Edges
Chalk's research, delivered to you at halftime.
Chalk publishes research-backed betting edges — patterns with statistically validated performance across multiple out-of-sample seasons. These appear as filter chips on your dashboard toolbar. Enable SMS on any edge and Chalk texts you at halftime when a live game matches.
Power4/Major conferences, home trailing 6-15 at half, model less optimistic than market.
Home trailing 6-15 at half, model more optimistic than market.
Each edge includes the thesis, backtest results, and exact filter parameters. New edges are published as Chalk's research validates them. Past performance does not guarantee future results.
How Chalk Works
Your game-day workflow in four steps.
Get the Alert
Chalk's published edges fire at halftime. You receive an SMS with the game, the matched pattern, and the model-vs-market divergence. Tap the link to open the game page and evaluate. No setup required beyond toggling on the edges you want.
Evaluate the Signal
The game detail page shows everything you need: regime classification, model prediction vs. market line, confidence interval, conference tier, and signal direction. Decide whether this game fits your thesis. Track it with one tap.
Manage Your Position
Cover Dynamics tracks your cover margin in real time. The trajectory chart shows momentum. The hedge calculator sizes exit trades. Fair cashout valuation tells you whether your sportsbook's offer is fair. At the 9-minute checkpoint, Chalk assesses whether your position is on track.
Review and Learn
After the game, the recap page shows the full story: how cover probability evolved, key inflection points, and whether the edge played out. Your paper book logs the result and tracks your season record.
Everything above, plus the tools to build your own strategies. Strategy Lab backtests any combination of filters across 4+ seasons. The full filter builder unlocks all fields with unlimited triggers and 25 SMS/day. Position Greeks (θ, Δ, Γ, ν) add options-style analysis to Cover Dynamics. Compare tiers →
Getting Started
First-time setup and navigation.
Tour of the Dashboard
The Dashboard is the main screen. It shows every active and upcoming game as a card with live scores, status badges, and betting signals.
- League toggle (top toolbar): Switch between All, CBB (college basketball), and NBA.
- Date navigation: Browse today's games or select a past date to review.
- View modes: Grid view (larger cards) or List view (compact rows).
- Halftime filter: Show only games at or past halftime — useful when signals have arrived and you want to focus on actionable games.
- Quick-filter dropdowns (all tiers): Narrow by Conference Tier, Regime, or Signal direction from the toolbar.
- Filter Builder (Pro+): Open the custom filter drawer to define complex conditions — numeric ranges, categorical filters, and field comparisons.
- Saved filters and strategies (Pro+/Syndicate): Select a saved filter or named strategy from the dropdown. Matching games are highlighted or filtered based on your selected mode.
Understanding Signal Badges
After a game reaches halftime, signal badges appear on game cards. These encode the key decision inputs at a glance.
Classifies the game based on the home team's halftime deficit or lead. TRAIL 6-15 is historically the highest-edge regime. LEAD is the lowest.
Direction of the model-vs-market divergence. BULL means the model is more optimistic on the home team than the market. BEAR means less optimistic. Pro+ users see the divergence magnitude in points.
P4 = Power4 (ACC, Big Ten, Big 12, SEC). MAJ = Major (A10, Big East, WCC). MM = Mid-Major.
A gold star appears when a game matches the target pattern — Power4/Major conference, trailing regime, and BEARISH signal. This combination has produced 72.1% ATS across four out-of-sample seasons (n=283). Past performance does not guarantee future results.
Adding Games to Your Tracked List
Hover over any game card and click the plus icon in the top-right corner. The game is added to your Tracked list (accessible from the sidebar). Tracked games feed into Cover Dynamics for live position management during the second half.
To remove a game from tracking, click the check icon that replaces the plus icon, or dismiss it from the Cover Dynamics page.
Reading the Game Detail Page
Click any game card to open the full game detail view. This page includes:
- Header card: Teams, live scores (with flash animation on score changes), game clock, and status badges.
- Key Stats: Summary cards showing the latest model prediction, confidence interval, and halftime insight when available.
- Trajectory chart: Visual history of model predictions over the course of the game. Updates in real time during live games.
- Predictions table: All active models' latest predictions with point estimates and confidence intervals.
- Halftime Insight (at/after halftime): Regime classification, model-vs-market divergence, confidence interval, and signal direction.
- Betting Panel: Market line, model implied 2H margin, and divergence.
Feature Guides
Detailed walkthroughs for each product area.
Research and customization tools for quantitative bettors.
Concepts & Terminology
Reference definitions for the metrics and concepts used throughout Chalk.
Models
Chalk combines proprietary deep learning and statistical models into an ensemble. Each model takes the current game state and produces a prediction of the final margin with calibrated uncertainty bounds.
- Deep learning models: Trained on 60,000+ historical games of play-by-play data. They process the sequence of game events to predict the final spread. These models use quantile regression to produce calibrated confidence intervals rather than single point estimates.
- Statistical models: Mathematical models that apply stochastic process theory to score dynamics. They capture effects like mean-reversion — the tendency for large halftime leads to shrink — and the changing volatility of scoring runs.
- Halftime snapshot: A specialized model that locks in at halftime and predicts the final margin based solely on first-half performance. It does not update during the second half, providing a stable reference for how the game evolves versus halftime expectations.
Ensemble approach: No single model is dramatically better than the others. The edge comes from combining multiple approaches that capture different aspects of game dynamics. Chalk uses this ensemble to compare against the live market line and surface opportunities in Strategy Lab.
Metrics
- MAE (Mean Absolute Error): Average magnitude of prediction errors, regardless of direction. "On average, how many points off is the prediction?" Chalk's ensemble achieves ~11 points MAE. For context, the naive baseline (predicting halftime margin carries forward unchanged) achieves ~12.5 points MAE — the improvement over baseline is real but modest.
- Confidence intervals: Every prediction includes uncertainty bounds. The 68% CI is the "typical" uncertainty range (actual result falls within roughly 68% of the time). The 95% CI captures almost all outcomes. Chalk's intervals are calibrated: observed coverage closely matches the target percentages.
- Spread convention (away - home): All margins in Chalk use the away minus home convention. A negative value means the home team is winning or favored. A positive value means the away team is winning or favored. Example: a halftime margin of +9 means the away team leads by 9 (home is trailing by 9).
Regime
Games are classified into regimes based on the home team's halftime position:
| Regime | HT Margin (A-H) | Meaning |
|---|---|---|
| TRAIL 16+ | > +15 | Home down 16+ pts |
| TRAIL 6-15 | +6 to +15 | Home down 6-15 pts |
| CLOSE | -5 to +5 | Within 5 pts either way |
| LEAD 6-15 | -15 to -6 | Home up 6-15 pts |
| LEAD 16+ | < -15 | Home up 16+ pts |
Historically, the trailing regime (home down 6-15) shows the strongest edge for home team second-half performance. The leading regime shows the weakest.
Signal (BULLISH / BEARISH)
The signal compares Chalk's model prediction for the second half against the live market line.
model_implied_2h < market_2h_line.model_implied_2h > market_2h_line.The divergence magnitude (in points) indicates how far apart the model and market are. Larger divergence = stronger signal.
Target Pattern
The target pattern is the intersection of three conditions:
This combination has produced 72.1% ATS across four out-of-sample seasons (n=283). The sample spans 2022-23 through 2025-26, with each season validated on data the models had never seen during training.
It misses approximately 30% of the time. Individual weeks and months will have losing streaks. The edge is statistical, not deterministic.
Past performance does not guarantee future results.
Frequently Asked Questions
Important Information
Chalk provides statistical analysis and modeling tools for informational and educational purposes. It does not provide betting advice, recommendations, or guarantees of any outcome.
All published performance statistics (including 72.1% ATS target pattern, 58.4% trailing ATS) are calculated from out-of-sample historical data across four college basketball seasons (2022-26). Trailing ATS is adjusted for line slippage between exchange and retail sportsbook pricing. These results reflect past performance and do not guarantee future results. Individual months and seasons will have losing periods.
The target pattern (n=283) misses approximately 27% of the time. Model predictions carry uncertainty reflected in the published confidence intervals. Spreads and margins use the away-minus-home convention throughout.
If you choose to use this information in connection with sports wagering, do so responsibly and in accordance with applicable laws in your jurisdiction.
Still have questions?
Contact us for support or feedback.
support@chalk.is