How Krok Odds calculates edge

Data sources

Core formulas

• Implied probability
  p = 1 / decimal_odds
• True probability (margin removed)
  p_true(i) = p(i) / Σ p(j)
• Expected value
  EV% = (p_true × (odds − 1) − (1 − p_true)) × 100
• Arbitrage detection
  arb when Σ (1 / best_odds_i) < 1.00
• Kelly criterion (quarter)
  f* = 0.25 × edge / odds_decimal
• Closing line value
  CLV% = (close_prob × take_odds − 1) × 100

What we will not do

• Fabricate true probability. If no sharp baseline exists, EV % is hidden — not estimated.
• Inflate edges with stale prices. Markets older than 10 minutes from event start are flagged stale; older than 24h are excluded.
• Cherry-pick history. Track-record buckets need ≥5 settled decisions before they display.
• Hide a losing model. Wins, losses and pushes are tracked publicly per sport with bucketed confidence breakdowns.

FAQ

• How does Krok Odds generate its tips and predictions?

  Every pick starts with a market-consensus baseline — real-time odds from 100+ Australian bookmakers, de-vigged and averaged to produce a fair-line probability. On top of that sits an ensemble of statistical models (rating systems, form analysis, head-to-head records, venue factors) that each independently assess the game. A third layer applies contextual judgement — injury news, weather, travel fatigue, motivation — and either confirms or overrides the math when evidence is strong enough. The final pick is a blend of all three layers, weighted by each model's tracked accuracy.

• What is the "sharp line" used for EV calculations?

  Our sharp baseline comes from the lowest-margin, highest-volume markets globally — operators that price close to true probability. When that's unavailable, we blend the licensed AU exchange (post-commission) with the sharpest international reference we collect. If neither exists, EV is hidden rather than guessed.

• How does Krok Odds remove the bookmaker margin?

  For standard 2-way and 3-way markets we proportionally strip the overround: each implied probability is divided by the sum of all implied probabilities. For player props and longshot markets we use a method that better handles favourite-longshot bias.

• How do weather and venue factors affect the predictions?

  For outdoor sports (AFL, NRL, NFL, soccer), we pull live weather observations at each match venue — rainfall, wind speed, temperature. Wet conditions suppress scoring in AFL and NFL. Strong winds affect kicking accuracy in AFL and passing in NFL. These feed into our scoring projections alongside the statistical models.

• What is closing line value (CLV) and why does it matter?

  CLV measures whether our pick was taken at a better price than the final odds at kickoff. Consistently beating the closing line is the single strongest predictor of long-term profitability — it means our model sees value the market hasn't priced yet. We track CLV publicly per sport and confidence level.

• How does the model handle injuries and team news?

  We ingest live injury feeds and cross-reference them against player usage data. When a key player is ruled out, we model the cascading effect on teammates — usage redistribution, role changes, and adjusted projections. This is particularly impactful for NBA player props, where a star's absence can shift 2-3 teammates' projections significantly.

• How often are odds and predictions refreshed?

  AU corporate odds refresh every 60–300 seconds depending on proximity to event start. In-play markets refresh every 30–60 seconds. Our models re-run on every odds tick, and the AI analysis layer refreshes hourly with the latest news and injury data.

• Does Krok Odds use prediction markets like Polymarket or Kalshi?

  Yes. We blend probabilities from global prediction markets (Polymarket, Kalshi) into our consensus model. These markets aggregate thousands of bettors worldwide and often move before traditional bookmakers — they're a useful signal, especially for sports with thinner AU book coverage.

• What is the limitation period for tracked bets?

  CLV and ROI tracking retain individual bet records for 24 months. Aggregate model performance (win rate by sport, by confidence bucket) is retained indefinitely.