METHODOLOGY
Nothing hidden.
Here's exactly how Lytic calculates.
Transparency is the foundation of a credible analytics tool. Below we describe — without simplification — every algorithm we use, the assumptions it relies on and where it has known weaknesses.
1. Devigging (margin removal)
Bookmakers build a margin ("vig") into their odds — the sum of all implied probabilities for a market is always > 1. Before we can use odds as probability estimates, the margin must be removed.
Lytic uses two different devig models depending on the market type:
Power devig — two-way markets
Used for validated two-outcome markets such as tennis match winner and BTTS. Markets with pushes or overlapping outcomes require separate settlement-aware adapters before they can be recommended.
The method finds the exponent k such that the sum of odds values raised to the power k equals exactly 1:
The solution provides an exponential compression of implied probabilities that removes the margin mathematically without making assumptions about how the margin is distributed across outcomes.
Shin devig — three-way markets
Used for markets with three outcomes: football 1X2 (home/draw/away).
The Shin model (Shin, 1991) accounts for the fact that part of the bookmaker margin in three-way markets is caused by insider trading friction — insiders with an information advantage systematically inflate the margin.
The model solves for the insider friction parameter z via binary search and corrects the implied probabilities based on the estimated friction:
where q_i = 1/odds_i, Q = Σ q_i (the raw sum > 1), and z is solved iteratively until Σ p_i = 1.
Shin devig is more accurate for football 1X2 than power devig because it explicitly models the asymmetry created by insider trading.
2. Probability estimation
Once the margin is removed, we need to choose which reference gives the best estimate of true probability. Lytic uses a priority chain:
Pinnacle (no-vig)
The tightest margins in the world (~1–2%). Closest to true probability — our sole sharp reference.
Sharp consensus (legacy)
Retained for historical records only; it is not assigned while Pinnacle is the sole active sharp reference.
Recreational median
Median across all available bookmakers. Lowest priority.
Lytic always uses the highest available source in the chain. If Pinnacle does not offer the market, we fall back conservatively to the broader market median. Legacy sharp-consensus labels are retained for old records only while Pinnacle is the only active sharp reference.
EV thresholds per reference source
The weaker the reference source, the higher the EV requirement before Lytic flags a bet as value. A weak reference means more uncertainty in the probability estimate — the error margin must be covered by a larger edge.
| Reference source | Min. EV |
|---|---|
| Pinnacle (no-vig) | +1.0% |
| Sharp consensus | +6.0% |
| Recreational median | +6.0% |
The principle: the weaker the reference, the higher the EV requirement before Lytic flags a bet as value.
3. Expected Value (EV)
EV measures how much a bet is expected to be worth relative to its price. A positive EV means the odds pay more than the probability warrants — a negative EV means the opposite.
Example
True probability = 55% | Offered odds = 2.00
EV = 0.55 × 2.00 − 1 = +10%
With $100 staked, this bet expects a net result of +$10 on average. Individual outcomes are binary — either +$100 (win) or −$100 (loss).
4. Profitability and NO BET
Positive headline EV is only the first filter. Before a signal can be recommended, Lytic subtracts a market-specific probability buffer and execution costs, then requires positive expected logarithmic bankroll growth. This prevents a noisy longshot or thin market from outranking a smaller but more reliable edge.
The policy changes by market family because liquidity, settlement complexity, price movement and model error are not equal across markets. The current validated limits are read from the same policy object used by the scanner:
| Market family | Min. net EV | Min. books | Max. odds |
|---|---|---|---|
| Match winner, spreads, qualify | +1.25% | 3 | 5.00 |
| Totals, team totals, BTTS | +1.75% | 3 | 5.00 |
| Validated corners and cards | +3.00% | 3 | 6.00 |
| Validated player props | +4.00% | 3 | 8.00 |
| Validated futures | +5.00% | 3 | 25.00 |
| Exchange quotes | +1.50% | 1 | 20.00 |
| Validated live markets | +3.00% | 2 | 8.00 |
| Executable parlays | +6.00% | 1 | 30.00 |
Unknown markets fail closed. Quarter-lines, push-capable markets, exchanges and parlays are not treated as ordinary two-way bets: Lytic returns NO BET unless the required settlement or executable-price evidence is available. Approved signals are ranked by conservative profitability and expected log growth—not hit rate alone.
5. Kelly Criterion
The Kelly Criterion (Kelly, 1956) calculates the optimal fraction of the bankroll to stake to target long-term growth. It cannot eliminate loss or ruin risk, especially when the probability estimate is wrong.
fFraction of bankroll to stakebNet odds — decimal odds minus 1 (odds 2.00 → b = 1.00)pEstimated probability of winningq1 − p (probability of losing)
Quarter-Kelly (Lytic default)
Lytic uses 25% of full Kelly as the default setting. Full Kelly is mathematically optimal given a precise edge estimate — but in practice, edge estimates carry uncertainty, and full Kelly produces severe drawdowns when estimates are overstated.
Default
25%
Quarter-Kelly
Maximum
50%
Half-Kelly cap
The Kelly fraction is hard-capped at 50% of full Kelly — regardless of how high a bet's EV appears. The final stake is also capped by the user's maximum-stake setting (5% of bankroll by default) and revalidated atomically when the bet is saved. The Kelly stake adjusts automatically as the bankroll grows or shrinks: when the bankroll increases, stakes increase proportionally, and vice versa.
6. Closing Line Value (CLV)
CLV measures whether you got a better price than the final market price before the match started ("closing line"). It is a leading measure of price quality that is less noisy than short-run win rate or ROI, but it is not a guarantee or standalone proof of future profit.
Example
Placed odds = 2.10 | Pinnacle closing no-vig = 2.00
CLV = (2.10 / 2.00 − 1) × 100% = +5.0%
Why CLV matters more than short-term ROI
ROI measures outcomes — which are largely driven by variance over small samples. CLV measures the price you paid compared to the market's best estimate at closing time. It is a measure of information advantage, not luck.
Rule of thumb
Sustained positive average CLV over a large sample is evidence that you consistently obtain better prices. Review it together with calibration, sample size, execution slippage and realised ROI.
Lytic captures CLV automatically when a compatible closing reference is available. Pinnacle no-vig is preferred; a sanity-checked median market price is used as a fallback.
Questions about the methodology?