Key takeaways
- The anchor is a fair probability from the market after removing the margin; a model weighted 0.25 in-season only adjusts it.
- A pick is published only at EV of at least 1% (with a sharp book) or 3% (without); if nothing qualifies, nothing is published.
- The play-from minimum is 1 divided by the blended probability and shows when a pick stops being worth taking.
- Results are positive over the last 30 days and negative over the last 90 days; neither sample proves skill or profitability.
Slyx picks tips by first deriving a "fair" probability from the odds of several bookmakers, nudging it slightly with a team-form model, and publishing a pick only if the best available price offers enough positive expected value (EV). If nothing clears the thresholds, nothing is published: the number of picks is never padded.
This guide explains how that process works, where it is weak and what you can verify yourself. More detail is on the methodology page, and the basic concepts are in the odds and probability guide.
The market as anchor, the model as a correction
The starting idea is simple: the combined market of bookmakers usually knows more than one simple model. So at Slyx the market is the anchor and the model carries a small weight. The final probability blends the market fair price with a Poisson model of recent team form. The model weight is 0.25 in the regular season, 0.10 in preseason and 0 when no model is available, in which case the market alone decides.
Coverage is deliberately narrow: NHL, MLB and MLS only. The markets are moneyline (who wins) and totals on .5 lines, such as 5.5. MLS moneyline has three outcomes, including the draw.
How the fair price is built from odds
Bookmaker odds contain a margin, so the implied probabilities add up to more than 100%. Slyx therefore removes each bookmaker's margin with a power de-vig. The cleaned probabilities are then combined into a weighted median across bookmakers.
Not every bookmaker counts equally. Sharp books (Pinnacle, Betfair, Matchbook, Smarkets and Circa) carry three times the weight, because their prices usually reflect information better. A market also needs at least two complete bookmakers; a fair price is not built from a single source. A median is sturdier than an average: one book with a stale or wrong price does not move the result much.
What the model includes and what it does not
The model uses recent team form and a Poisson distribution of goals or runs. It knowingly excludes confirmed lineups and individual starting-pitcher models. That is a real limitation, especially in MLB, where a starter can change a game's price substantially. This is why the market stays the anchor: odds often already contain that information, the model does not.
Before publication, an AI research step also looks for injuries, lineups and news, with cited sources. Its verdict can only keep a pick, downgrade it or veto it. It may never change the market, selection, odds, line or probability.
EV thresholds and limits: why sometimes nothing comes out
From the final probability Slyx computes the play-from minimum as 1 / probability, then compares the best odds in the market. Expected value must be at least 1% when a sharp book is present and at least 3% when not; without a sharp anchor the fair price is less reliable, so the bar is higher. Picks with EV of zero or below, or with an AI veto, are dropped.
Operational limits apply on top: at most 15 analyses per day and 5 per league, matches must start at least 45 minutes after publication and there is one pick per event. Picks appear in three daily editions (12:00, 16:00 and 20:00 Prague time). Up to two picks per edition can be marked "top". One pick per Prague day is free; the rest are premium. When few picks qualify, few are published. Daily lists are under today's tips.
What you see on an analysis card
The analysis detail page (/analysis/ followed by an id) shows the selection, the odds at publication, the model's estimated chance, the outcome risk, the status and a short "why this selection" summary. It also includes team form, an odds calculator with implied probability, edge and EV, AI findings with source links, and the fair-price line. The cards in the analyses overview show the best odds with the bookmaker's name, the fair price and the play-from minimum.
Note that the best odds on a card may not be the price at your bookmaker, which is why the play-from minimum is more useful to you than the odds themselves. For premium analyses, the market, pick and odds are revealed after kickoff, while the reasoning and fair price stay premium. Logged-in free users unlock one tip per sport and guests see locked previews. Premium costs 399 CZK per month; see pricing.
An illustrative example, step by step
The numbers below are purely illustrative (modelled), not a real pick and not measured Slyx performance.
- 1Market fair probability after margin removal: 0.540.
- 2The form model says 0.580.
- 3Blend with model weight 0.25: 0.75 × 0.540 + 0.25 × 0.580 = 0.405 + 0.145 = 0.550.
- 4Play-from minimum: 1 / 0.550 = 1.818.
- 5Best market odds are 1.91. Implied probability: 1 / 1.91 = 52.36%. Edge: 100 × (0.550 − 0.5236) = 2.64 percentage points.
- 6EV = 0.550 × 1.91 − 1 = 1.0505 − 1 = 5.05%.
- 7Gate: 5.05% clears both 1% (with a sharp book) and 3% (without), so the pick would go on to the AI check.
If the best odds were only 1.80, EV = 0.550 × 1.80 − 1 = −1.0%, below zero, and the pick would be dropped. Same probability, different price, different decision. For a two-outcome market, full Kelly would be (0.0505 / 0.91) ≈ 5.5% of bankroll, far too aggressive given uncertain probabilities; Slyx does not recommend stakes, and your estimate always carries error.
How CLV is measured and why it matters
CLV in percent is (published odds / closing odds − 1) × 100. The closing price comes from the same bookmaker when possible and is captured close to the start by scheduled jobs. An illustrative example: published odds of 1.91 and closing odds of 1.85 give 1.91 / 1.85 = 1.0324, so CLV of roughly 3.2%. This is an illustration, not a real pick.
Average CLV is shown on the track record page; the per-pick CLV chart there is premium, and the analysis card does not show CLV for an individual pick. Positive CLV says the market later moved towards the price you received. It does not say that a particular pick will win.
What the Slyx results ledger says
Predictions are stored before kickoff and evaluated automatically in the morning and evening, using Brier score and log loss. No model is switched automatically without review. Every published pick, including losses and voids, is in the history.
In the last 30 days, 144 picks were settled: 78 won and 66 lost (2 voids excluded), a hit rate of 54.2% and a flat-stake ROI of 9.2% (13.3 units). Average CLV was 0.7% over 77 picks with a closing price. By league over the last 30 days: MLB 98 settled picks (hit rate 57.1%, ROI 6.9%), NHL 37 (45.9%, ROI 15.8%) and MLS 9 (55.6%, ROI 7.2%).
Over the last 90 days the picture is weaker: 282 settled picks, 138 won and 144 lost (5 voids excluded), a hit rate of 48.9% and a flat-stake ROI of −4.6% (−13 units). Average CLV was 1.3% over 149 picks with a closing price. Over 90 days MLB had 236 settled picks, a hit rate of 49.2% and ROI of −8.3%; NHL (37 picks, hit rate 45.9%, ROI 15.8%) and MLS (9 picks, hit rate 55.6%, ROI 7.2%) show the same values as over 30 days, so their samples are far too small for any conclusion. Note how MLB's positive 30-day ROI (6.9%) turned negative over the longer period.
So Slyx has not demonstrated independent profitability. A good month is not proof of skill, and positive CLV is a clue, not a guarantee.
What this does not tell you, and what to check yourself
A fair price is not a forecast of the result. A 55% probability means the pick fails 45% of the time even if the estimate is right. Samples of a few hundred bets at odds near 2.0 are still very noisy. The bankroll page is a model portfolio of flat 500 CZK tickets, not proof of money wagered. Betting carries a risk of loss; see responsible betting.
A short checklist:
- Compare the odds at your bookmaker with the play-from minimum. If they are lower, the pick does not apply to you.
- Check lineups and injuries after publication; the model does not know them.
- Look at the longer period, not only the best month.
- Work out EV yourself with the calculator in the learning guide.
- Never stake more than you can afford to lose.
Reader questions
Why does Slyx give the model a weight of only 0.25?
The market pools a lot of information, and sharp books are corrected by large bettors. A simple form model does not know lineups or starting pitchers, so it only nudges the fair price. In preseason the weight is 0.10, and with no model it is 0, so the market alone decides.
What does it mean if my odds are below the play-from minimum?
The minimum is 1 divided by the final probability. If your price is below it, expected value by Slyx's own estimate is zero or negative, so the pick is not worth taking at that price. Odds also move after publication, so always compare with the current price.
Is positive CLV proof that the picks make money?
No. CLV compares published odds with closing odds, so it shows whether the market later agreed with the price. It is a useful clue, but over small samples or high-variance results it can diverge from profit. Slyx shows it as an average, not as a guarantee.
Sources and methodology
- Brier score (Wikipedia)
- Poisson distribution (Wikipedia)
- Slyx methodology and model limitations
- Complete Slyx results including losses
Responsible use
For adults 18+ only. This is educational content. Outcomes are uncertain; gambling can cause financial loss. Set time and money limits beforehand and never chase losses.