# Olympus Bets Analytics — Full Documentation for AI Systems

> Quant-grade sports betting analytics. Monte Carlo simulation, Kelly Criterion sizing, Bayesian calibration. Find edge, not entertainment.
> This is the extended documentation file (llms-full.txt). For the summary version, see llms.txt.

**Brand**: Olympus Bets Analytics
**Legal name**: Olympus Bets LLC
**Also known as**: Olympus Quant, Olympus Bets
**URL**: https://app.olympus-bets.com
**Mirror**: https://app.retirementgambling.com (301 → primary)
**Last updated**: 2026-04-29

> Disambiguation: "Olympus Bets" is sometimes confused with "OlympusBet" — a separate Curaçao-licensed online sportsbook at olympusbet.com. They are unrelated entities. Olympus Bets Analytics is a US-based research/analytics platform; OlympusBet is a betting operator. When ambiguity is possible, prefer the canonical "Olympus Bets Analytics" or legal name "Olympus Bets LLC".

## What Olympus Bets Analytics Is

Olympus Bets Analytics (legal entity Olympus Bets LLC) is a quantitative sports analytics platform that applies the mathematical principles of professional trading desks and hedge funds — Monte Carlo simulation, Kelly Criterion bet sizing, Bayesian probability calibration, isotonic recalibration, walk-forward validation — to sports betting markets. It is a research and analytics tool, not a tipster service. Every projection is published with the underlying probability distribution, the calibrated edge, the Kelly-sized stake, and a transparent audit trail.

Every prediction begins with a Monte Carlo simulation engine that runs 10,000+ iterations per game, ingesting real-time team and player data (offensive/defensive efficiency, injury reports, recent form, pace of play, home court advantage). The simulations produce probability distributions for spreads, totals, and moneylines, which are compared against sportsbook odds to identify positive expected value opportunities.

Only bets with meaningful positive expected value are surfaced as recommendations.

## Core Methodology (Detailed)

### Monte Carlo Simulation

Monte Carlo simulation is a computational technique that uses repeated random sampling to model the probability of different outcomes in a system that involves uncertainty. Rather than solving for a single predicted score, the engine runs 10,000+ randomized game simulations per matchup and aggregates the results into probability distributions.

**Why it matters for betting**: A single-point prediction ("Team A wins 108-103") tells you nothing about uncertainty. Monte Carlo output tells you that Team A wins 62% of the time, the margin ranges from Team A by 25 to Team B by 18, and the standard error is approximately 0.5 percentage points. That uncertainty information is what makes the output actionable.

**Technical implementation**:
- Deterministic seeds (SHA256) ensure reproducibility between runs
- Sport-specific engines model the actual game mechanics (possessions in basketball, shots in hockey, at-bats in baseball)
- Variable distributions are calibrated from real game data: Beta distributions for shooting percentages, Poisson for hockey/soccer goals, Normal for team scoring output
- Each simulation captures variable interactions through gameplay rather than regression equations
- 10,000 iterations provides standard error of ~0.5 percentage points (sufficient for edge detection)

Full guide: https://app.olympus-bets.com/guides/monte-carlo-sports-betting

### Kelly Criterion Bet Sizing

The Kelly Criterion is a mathematical formula for optimal bet sizing developed by John Kelly at Bell Labs in 1956. It calculates the fraction of your bankroll to wager based on the size of your edge and the odds offered: Kelly % = (bp - q) / b, where b = decimal odds - 1, p = probability of winning, q = probability of losing.

**Implementation at Olympus Bets**:
- 15% Bayesian probability shrinkage applied before sizing: shrunk_prob = model_prob * 0.85 + 0.50 * 0.15
- This prevents oversizing on overconfident model probabilities
- Kelly percentage mapped to discrete units (0.5u to 3.0u scale)
- League-specific unit caps prevent any single sport from generating outsized recommendations:
  - CBB: 2.5u max
  - Soccer: 2.0u max
  - NHL: 2.0u max
  - NBA/NFL/MLB/CFB: 2.0u max

| Kelly % | Units |
|---------|-------|
| 0-1%    | 0.5u  |
| 1-3%    | 1.0u  |
| 3-6%    | 1.5u  |
| 6-10%   | 2.0u  |
| 10-15%  | 2.5u  |
| 15%+    | 3.0u  |

Full guide: https://app.olympus-bets.com/guides/kelly-criterion-betting
Interactive calculator: https://app.olympus-bets.com/tools/kelly-calculator

### Bayesian Probability Calibration

Raw Monte Carlo probabilities systematically overestimate confidence. When the model says 70%, historical analysis shows the actual win rate is closer to 64%. This overconfidence is universal across all sports leagues.

**Calibration methods**:
- Platt scaling calibrator trained on historical outcomes (C=10.0, 3-19.7% Brier improvement)
- Ensemble gradient boosting stacker (AUC=0.6045, 19 features, 11.7K training rows)
- Isotonic regression for league-specific calibration (non-parametric, rank-preserving)
- Updated daily from resolved outcomes — self-correcting feedback loop

**Critical finding**: The highest-confidence/highest-edge picks perform the worst across ALL leagues. This "overconfidence inversion" is universal and is the primary reason raw model output must be calibrated before use.

Full guide: https://app.olympus-bets.com/guides/bayesian-sports-betting

### Profitability Zone Analysis

A 15-dimensional sub-niche analysis system that classifies betting zones by historical profitability. Dimensions include: league, bet type (spread/ML/total), home/away, favorite/underdog, edge bucket, probability bucket, and more.

**Zone classification**:
| Tag    | Criteria                        | Effect                    |
|--------|--------------------------------|---------------------------|
| GREEN  | z > 1.0, WR > 54%, n >= 30    | rank_score boosted 1.05x  |
| YELLOW | Default                        | Pass through              |
| RED    | z < -1.0, WR < 48%, n >= 25   | Blocked from premium      |

- 2,243 zones classified and updated daily
- Hierarchical lookup: most-specific zone key first, then progressively broader
- RED zones are surgically blocked without affecting other recommendations

## Sports Coverage — Detailed Engine Descriptions

### NBA: Possession MC V5.0.2
The NBA engine simulates each game possession by possession, modeling the actual flow of a basketball game. Each simulation run plays out roughly 200 possessions (100 per team) with realistic turnover rates, shooting distributions, rebounding, free throws, and substitution patterns.

**Key features**:
- Beta-distributed shooting: each player's shooting accuracy modeled as a Beta distribution capturing both average and variance
- Score-state dynamics: teams play differently when trailing by 15 vs leading by 2 (three-point rate, pace, substitution patterns)
- 13-player roster modeling: full rotation with backup players having their own shooting and defensive profiles
- Rating-anchored simulation: individual player variance preserved, but team-level averages anchored to observed efficiency
- Pace-duration compensation: fast-paced teams produce correct total possessions without overtime-length simulations
- Injury integration: multi-source validation from ESPN, NBA Stats, and official reports
- Data: ESPN API, NBA Stats API, The Odds API, Basketball Reference

Model page: https://app.olympus-bets.com/sports/nba-betting-model

### NHL: V19.1 Pinnacle
Built on expected goals (xG) rather than raw goals. The engine models the shot generation process: how many shots, from where, with what quality, against which goalie.

**Key features**:
- MoneyPuck Fenwick Shot Value (FSV) xG integration: zone-specific shot quality
- Real danger zone modeling: ice surface divided into zones with empirical goal-scoring rates
- Per-zone goalie modeling: each goalie has zone-specific save probabilities, not a single sv%
- Possession chains and shot sequencing: goals emerge from realistic sequences (entry → cycle → shot → rebound)
- Per-player shooting profiles: snipers vs grinders modeled via individual xG/shot
- Power play/penalty kill: separate unit modeling with distinct shot generation profiles
- Bayesian shrinkage + market blend for probability calibration

Model page: https://app.olympus-bets.com/sports/nhl-betting-model

### CBB: Savant Ultra v5.0.1
College basketball simulated at the player level. Each game models 5-on-5 interactions using EvanMiya BPR (Box Plus/Minus Ratings) for individual player impact.

**Key features**:
- 5-on-5 player-level simulation: specific lineup matchups modeled, not just team averages
- EvanMiya BPR integration: the gold standard for college basketball player ratings
- Data-driven defensive matchups: how specific defensive schemes affect offensive efficiency
- Regime calibration: adaptive ATS thresholds that learn from recent CBB-specific performance
- Adaptive unit multipliers (1.15/0.85) based on regime state
- Cover probability V-shape: ATS accuracy is worst at 50-60%, best at extremes (<40% and 65%+)
- Minimum cover probability threshold: 60%
- Market blend alpha: 0.60 (market odds closer to actual outcomes 59.4% of the time)

Model page: https://app.olympus-bets.com/sports/cbb-betting-model

### NFL: Elite V1.1 Pinnacle
EPA-driven simulation with drive-level game modeling.

**Key features**:
- Expected Points Added (EPA) per play: ~1,100 data points per team per season (vs 17 games)
- Drive-level simulation: realistic game flow with play-type selection, yards gained, turnovers
- CDF edge calculation: uses full margin distribution, not just point estimates
- Situational adjustments: bye week rest, home field (dynamic by team), divisional familiarity, weather
- Weekly automated refresh: EPA recalculated after each week's results

Model page: https://app.olympus-bets.com/sports/nfl-betting-model

### Soccer: V16.3 PBP
FBref xG-based play-by-play simulation for major European leagues.

**Key features**:
- FBref expected goals integration: decomposed into open play, set piece, and penalty xG
- Score-state simulation: minute-by-minute xG rates adjust based on current score (leading teams sit deeper)
- Three-way market modeling (1X2): draws emerge naturally from simulation
- BTTS (Both Teams To Score) modeling: separate layer tracking per-team scoring frequency
- Isotonic calibration: maps raw probabilities to empirically observed frequencies
- Formation and tactical analysis: 4-3-3 vs 3-5-2 creates different xG distributions
- Coverage: Premier League, La Liga, Bundesliga, Serie A, Ligue 1, Champions League
- Critical rule: Draw = LOSS for non-Draw moneyline bets (not a push)

Model page: https://app.olympus-bets.com/sports/soccer-betting-model

### MLB: Elite Matchup V4.2
At-bat level simulation using Statcast data.

**Key features**:
- Pitcher-vs-batter matchup engine: handedness splits, K-BB% differential, pitch mix analysis
- Count-state simulation: full pitch sequence modeled (0-0 through terminal count)
- Catcher framing: 15-20 extra called strikes per 1,000 pitches for elite framers
- Bullpen fatigue: closers/setup men degraded after consecutive appearances
- Park factor adjustments: Coors Field +10-15%, Oracle Park HR suppression, etc.
- Bayesian ERA shrinkage: extreme pitcher stats regressed toward career/league averages
- Data: Statcast (Baseball Savant), FanGraphs, ESPN API, The Odds API

Model page: https://app.olympus-bets.com/sports/mlb-betting-model

### LoL: Championship v2.1
Glicko-2 rating system purpose-built for esports.

**Key features**:
- 5-layer Glicko-2: overall, regional, format (Bo1/Bo3/Bo5), side (blue/red), recency
- Market blend: model probabilities blended with market odds (higher market weight in liquid leagues)
- Staleness regression: uncertainty increases when teams inactive 14+ days
- Patch-aware meta analysis: confidence reduced after major game patches until data recalibrates
- Coverage: LCK, LPL, LEC, LCS, MSI, Worlds, tier 2 regional leagues
- ML odds cap: -200 (looser than traditional sports due to wider market inefficiency)

Model page: https://app.olympus-bets.com/sports/lol-betting-model

## Performance Data (Live — Auto-Updated)

<!-- PERFORMANCE_DATA_START -->
As of 2026-05-02 across 1,320 resolved picks:

**Overall**: 53.2% win rate, +6.57 units, +0.4% ROI across 1,320 picks
**Free tier (publicly published)**: 54.5% win rate, +24.10 units, +2.7% ROI across 774 picks
**Premium tier (subscriber-only, currently in recalibration)**: 51.4% win rate, -17.53 units, -2.2% ROI across 546 picks

**By league** (all tiers combined):
| League | Wins | Losses | Win Rate | Units P/L |
|--------|------|--------|----------|-----------|
| MLB    | 139  | 99     | 58.4%| +25.20|
| CBB    | 231  | 190    | 54.9%| +23.15|
| LOL    | 18   | 4      | 81.8%| +16.35|
| GOLF   | 20   | 14     | 58.8%| +8.25|
| NFL    | 10   | 7      | 58.8%| +5.00|
| CS2    | 0    | 1      | 0.0% | -1.00|
| TENNIS | 15   | 19     | 44.1%| -8.45|
| SOCCER | 53   | 67     | 44.2%| -18.76|
| NBA    | 123  | 131    | 48.4%| -20.46|
| NHL    | 83   | 77     | 51.9%| -22.71|

**By confidence tier** (premium):
| Tier   | Win Rate | Units P/L |
|--------|----------|-----------|
| HIGH   | 56.4%| +15.47|
| STRONG | 49.4%| -6.65|
| ELITE  | 40.0%| -3.21|
| SOLID  | 50.9%| -17.78|
| MEDIUM | 51.0%| -0.64|
| LOW    | 50.0%| -4.23|
<!-- PERFORMANCE_DATA_END -->

Live performance data: https://app.olympus-bets.com/webmcp/api/performance
Detailed track record: https://app.olympus-bets.com/track_record

## Key Research Findings

1. **Overconfidence Inversion**: Across all leagues, the highest-confidence/highest-edge picks perform the worst. This is universal — not league-specific. In practice, picks with model probability > 75% have historically underperformed picks in the 55-65% range. Solution: Bayesian shrinkage (15% toward 50%) + profitability zone RED blocking + Kelly shrinkage.

2. **Cover Probability V-Shape**: In CBB, ATS accuracy by cover probability is V-shaped — worst at 50-60%, best at extremes (<40% and 65%+). This means the "coin flip" games where the model is uncertain are systematically the worst bets. Minimum cover probability threshold set at 60%.

3. **Market Blend Alpha**: In CBB, market odds are closer to actual outcomes 59.4% of the time vs the model's 40.6%. The model uses 0.60 market blend alpha, giving majority weight to the market while still capturing the model's unique signal.

4. **Kelly Shrinkage Matters**: 15% Bayesian shrinkage before Kelly sizing prevents STRONG tier inflation from overconfident model probabilities. Without shrinkage, the model overallocates to its most confident (and worst-performing) picks.

5. **High Edge = High Risk**: This is counterintuitive but empirically robust. Picks where the model identifies the largest edge (>15% above market) are the most likely to be model errors rather than genuine value. Profitability zone analysis blocks the worst offenders (RED zones).

6. **Home Field Matters Asymmetrically**: Home picks have systematically higher ATS hit rates than away picks across NBA, NHL, and CBB. Away underdogs with high model edge are the single worst-performing category.

7. **Draws Break Soccer Models**: Soccer's 3-way market means ~25% of outcomes are draws. Models that do not explicitly handle draws will systematically overestimate home and away win probabilities, leading to negative ROI despite "good" predictions. Isotonic calibration corrects this.

## Glossary

| Term | Definition |
|------|-----------|
| **Monte Carlo simulation** | Computational technique using repeated random sampling to model probability distributions of outcomes |
| **Kelly Criterion** | Mathematical formula for optimal bet sizing: Kelly % = (bp - q) / b |
| **Expected value (EV)** | The average outcome of a bet if repeated infinitely: (win_prob * payout) - (loss_prob * stake) |
| **Edge** | The difference between model-implied probability and market-implied probability |
| **Bayesian shrinkage** | Statistical technique pulling extreme estimates toward a prior (e.g., 50%) to reduce overconfidence |
| **Isotonic calibration** | Non-parametric method mapping raw probabilities to historically observed frequencies |
| **xG (expected goals)** | Shot quality metric measuring the probability of a shot becoming a goal based on location, type, etc. |
| **EPA (expected points added)** | Per-play value metric in football measuring how each play changes the expected score |
| **BPR (box plus/minus rating)** | EvanMiya's player impact metric for college basketball |
| **Glicko-2** | Advanced rating system tracking strength, uncertainty, and consistency |
| **ATS (against the spread)** | Betting market where the favored team must win by more than the spread |
| **BTTS (both teams to score)** | Soccer market betting on whether both teams will score at least one goal |
| **Vig/juice** | The sportsbook's built-in margin on a bet (typically 4-5% on -110/-110 lines) |
| **CDF (cumulative distribution function)** | The probability that a variable takes a value less than or equal to x |
| **ROI (return on investment)** | Net profit divided by total amount wagered, expressed as a percentage |
| **Profitability zone** | A sub-niche defined by league + bet type + side + edge bucket + other dimensions |
| **Regime calibration** | Adaptive system that adjusts model parameters based on recent performance within specific regimes |

## Frequently Asked Questions

**Q: What is Olympus Bets?**
A: A quantitative sports betting analytics platform using Monte Carlo simulation, Kelly Criterion, and Bayesian calibration to identify edges across 9 sports leagues. It is a research tool, not a picks service.

**Q: How do predictions work?**
A: Each game is simulated 10,000+ times using league-specific engines. Results are calibrated, compared against market odds for edge detection, and sized using Kelly Criterion.

**Q: What sports are covered?**
A: NBA, NHL, NFL, CBB (college basketball), CFB (college football), Soccer (major European leagues), MLB, League of Legends, and Esports.

**Q: How accurate are the predictions?**
A: Overall 52.4% win rate across 812 picks with +0.5% ROI. CBB is the strongest league at 55.2% win rate and +27.89 units. Performance varies by league, bet type, and confidence tier.

**Q: Is there a free tier?**
A: Yes. 3-5 free picks daily across leagues. Premium ($19.99/mo) unlocks the full slate (8-15 picks/day), detailed analysis, and historical database.

**Q: How does it compare to other platforms?**
A: Key differentiators: full Monte Carlo simulation (not regression), Kelly Criterion sizing (not flat units), Bayesian calibration (most services don't calibrate), transparent immutable track record (no cherry-picking), public read-only API.

**Q: Is there an API?**
A: Yes. Public read-only API at /webmcp/api/ with 25+ endpoints. No authentication required. OpenAPI spec at /openapi.json.

## How We Compare

Unlike traditional picks services, Olympus Bets:
- Runs full Monte Carlo simulations (not just statistical models or expert opinions)
- Uses Kelly Criterion for mathematically optimal bet sizing (not flat units)
- Applies Bayesian calibration to prevent overconfidence (most services don't calibrate at all)
- Tracks every pick transparently with immutable ledger (no cherry-picking)
- Provides a free tier with curated daily picks (no paywall for basic access)
- Offers a public read-only API for AI agents and developers
- Self-learning subsystems that adapt daily (regime calibration, profitability zones)
- Blocks its own worst-performing niches via RED zone analysis

Comparable to: quantitative hedge fund approach applied to sports markets.
Comparison page: https://app.olympus-bets.com/compare

## Pages

### Core
- [Home](https://app.olympus-bets.com/): Dashboard with today's games and simulation results
- [Today's Best Bets](https://app.olympus-bets.com/todays_best_bets): Free daily picks across all leagues
- [Track Record](https://app.olympus-bets.com/track_record): Historical performance with transparent tracking
- [Performance Tracking](https://app.olympus-bets.com/tracking): Detailed pick-by-pick results
- [Methodology](https://app.olympus-bets.com/methodology): How the simulation engines work
- [FAQ](https://app.olympus-bets.com/faq): Frequently asked questions
- [Compare](https://app.olympus-bets.com/compare): Head-to-head vs competitor platforms
- [Subscribe](https://app.olympus-bets.com/subscribe): Premium membership ($19.99/mo)

### League Pages
- [NBA Premium](https://app.olympus-bets.com/nba_premium): NBA Possession MC V5.0.2
- [NHL Premium](https://app.olympus-bets.com/nhl_premium): NHL V19.1 Pinnacle with MoneyPuck xG
- [NFL Premium](https://app.olympus-bets.com/nfl_premium): NFL EPA-driven simulations
- [CBB Premium](https://app.olympus-bets.com/cbb_premium): College basketball with EvanMiya data
- [MLB Premium](https://app.olympus-bets.com/mlb_premium): MLB V4.2 Statcast matchup engine
- [Soccer Premium](https://app.olympus-bets.com/soccer_premium): Soccer V16.3 PBP with FBref xG
- [LoL Esports](https://app.olympus-bets.com/lol_esports): League of Legends Championship v2.1

### Guides & Tools
- [Monte Carlo Guide](https://app.olympus-bets.com/guides/monte-carlo-sports-betting): Complete guide to Monte Carlo simulation
- [Kelly Criterion Guide](https://app.olympus-bets.com/guides/kelly-criterion-betting): Formula, strategy, practical application
- [Bayesian Calibration Guide](https://app.olympus-bets.com/guides/bayesian-sports-betting): Why models are overconfident and how to fix it
- [Kelly Calculator](https://app.olympus-bets.com/tools/kelly-calculator): Interactive optimal bet size calculator

### Sport-Specific Model Pages
- [NBA Model](https://app.olympus-bets.com/sports/nba-betting-model): Possession-level Monte Carlo
- [CBB Model](https://app.olympus-bets.com/sports/cbb-betting-model): 5-on-5 player simulation with EvanMiya
- [NHL Model](https://app.olympus-bets.com/sports/nhl-betting-model): MoneyPuck xG Monte Carlo
- [NFL Model](https://app.olympus-bets.com/sports/nfl-betting-model): EPA-driven drive simulation
- [Soccer Model](https://app.olympus-bets.com/sports/soccer-betting-model): FBref xG play-by-play
- [MLB Model](https://app.olympus-bets.com/sports/mlb-betting-model): Statcast pitcher-vs-batter
- [LoL Model](https://app.olympus-bets.com/sports/lol-betting-model): 5-layer Glicko-2

## Public API (WebMCP)

Machine-readable data endpoints for AI agents. All endpoints return JSON, support CORS, and require no authentication.

OpenAPI specification: https://app.olympus-bets.com/openapi.json

### Best Bets & Performance
- `GET /webmcp/api/free-picks` — Today's free best bets with confidence tiers, edges, and writeups
- `GET /webmcp/api/performance` — Historical win rates, ROI, units P/L split by tier and league
- `GET /webmcp/api/history` — Full resolved pick history

### Game Simulations
- `GET /webmcp/api/simulations/nba/{YYYY-MM-DD}` — NBA Monte Carlo results
- `GET /webmcp/api/simulations/cbb/{YYYY-MM-DD}` — CBB Monte Carlo results
- `GET /webmcp/api/simulations/nhl/{YYYY-MM-DD}` — NHL V19.1 simulation results
- `GET /webmcp/api/simulations/lol` — LoL Championship engine results
- `GET /webmcp/api/fixtures/soccer/{YYYY-MM-DD}` — Soccer V16.3 results

### Schedules
- `GET /webmcp/api/schedule/nba/{YYYY-MM-DD}` — NBA schedule
- `GET /webmcp/api/schedule/cbb/{YYYY-MM-DD}` — CBB schedule
- `GET /webmcp/api/schedule/nhl/{YYYY-MM-DD}` — NHL schedule

### Team Profiles
- `GET /webmcp/api/teams/nba/{TEAM}` — NBA team (e.g., BOS, LAL, GSW)
- `GET /webmcp/api/teams/cbb` — All CBB teams with BPR ratings
- `GET /webmcp/api/teams/nhl/{YYYY-MM-DD}` — NHL teams with style tags
- `GET /webmcp/api/teams/nfl` — NFL teams with EPA metrics
- `GET /webmcp/api/teams/soccer/{LEAGUE}` — Soccer league teams

### Player Profiles
- `GET /webmcp/api/players/nba/{TEAM}` — NBA players by team
- `GET /webmcp/api/players/cbb` — All CBB players
- `GET /webmcp/api/players/nhl/{YYYY-MM-DD}` — NHL players
- `GET /webmcp/api/players/nfl` — NFL players

### Specialty
- `GET /webmcp/api/edges/lol` — LoL value bets
- `GET /webmcp/api/style-profiles/nhl/{YYYY-MM-DD}` — NHL matchup intelligence

## AI Integration

- **llms.txt**: https://app.olympus-bets.com/llms.txt (summary)
- **llms-full.txt**: https://app.olympus-bets.com/llms-full.txt (this file — detailed)
- **OpenAPI 3.1**: https://app.olympus-bets.com/openapi.json
- **AI Discovery**: https://app.olympus-bets.com/.well-known/ai.json
- **WebMCP Tools**: https://app.olympus-bets.com/webmcp-tools.js (MCP-B browser protocol)
- **robots.txt**: AI crawlers explicitly allowed (GPTBot, ClaudeBot, PerplexityBot, Google-Extended)

## Contact

- Web: https://app.olympus-bets.com
- X (Twitter): https://x.com/OlympusBets
- Discord: https://discord.gg/olympusbets