Key takeaways
- Check an analysis with three questions: where the information came from, what price the market gives and when the information was current.
- Implied probability is 1 / odds, and it is not a fair probability until the bookmaker's margin is removed.
- A short results sample proves nothing; over the last 90 days Slyx shows an ROI of -1.5% across 277 settled picks.
- The odds at publication differ from the closing odds, so the timestamp matters.
You can check a sports analysis with three questions: where did the information come from, what price does the market put on the outcome, and when was the information valid? If an analysis cannot answer any one of them, it is an opinion rather than an analysis.
Below we take each question in turn, work through one illustrative calculation and describe how Slyx handles this. We also say plainly what the method cannot tell you.
Source: where the information came from
The first thing we check is traceability. A claim about an injury, a lineup or form should point to a specific report, not a vague "it is said". At Slyx, an AI research step runs before publication, looking for injuries, lineups and news and listing cited sources. The AI verdict can only keep a pick, downgrade it or veto it; it never changes the market, selection, odds, line or probability.
On the sources page we list what we draw on: Sport.cz, iSport and official league sources. They are not partners and not model inputs. A good habit is to confirm any important claim directly against an official source such as the league's own site.
Three levels of reliability
When you read any analysis, sort its claims in your head. The firmest is an official announcement from a team or league, with a date. Weaker is a news article that cites such an announcement. Weakest is a claim with no source at all, such as "the star probably will not play". Only the first two should move your estimate. And always ask whether the report has gone stale: yesterday's lineup guarantees nothing about today's game.
Market: what the price says
Odds are a price from which a probability can be derived, but they include the bookmaker's margin. The implied probabilities of all outcomes therefore add up to more than 100%. Slyx removes each bookmaker's margin with a power de-vig and takes the fair probability as a weighted median across books. Sharp books (Pinnacle, Betfair, Matchbook, Smarkets, Circa) weigh 3x, and a market needs at least two complete bookmakers.
An illustrative picture of margin: if both outcomes of an evenly matched game are offered at 1.90, the implied probability of each is 1 / 1.90 = 52.63%. The total is 105.26%, so about 5.26% is margin, and the fair chance here is 50% for each side. The figure is invented purely for explanation. Real markets are rarely symmetric, and a power de-vig spreads the margin in a more subtle way than simple division, but the principle is the same.
We cover moneyline and totals on .5 lines; MLS moneyline has three outcomes including the draw. The final probability blends the market fair price with a Poisson model of recent form: the model weight is 0.25 in the regular season, 0.10 in preseason and 0 when no model is available. The market is the anchor. The model does not include confirmed lineups or individual starting-pitcher models, which is a limitation you should know about.
Time: when the information was valid
Odds and news go stale. A price that held at noon may not hold in the evening, and an injury announced after publication changes the picture. That is why picks go out in three daily editions, at 12:00, 16:00 and 20:00 Prague time, and a match must start at least 45 minutes after publication. The analysis page (the /analysis/ address followed by the analysis number) shows the odds at publication.
The page also shows the selection, the model's estimated chance, outcome risk, status, a "why this selection" summary, team form, an odds calculator with implied probability, edge and EV, AI findings with source links and the fair-price line. For premium analyses, the market, pick and odds are revealed after kickoff; the reasoning and fair price stay premium.
Timing is also the basis for measuring quality. CLV is calculated as (published odds / closing odds - 1) x 100, with the closing price taken from the same bookmaker where possible, captured close to the start by scheduled jobs. Average CLV is shown on /track-record; the per-pick CLV chart is premium, and we do not show per-pick CLV on the analysis card.
Worked example: one calculation step by step
The numbers below are purely illustrative; this is not a real pick and not measured performance. Suppose your estimated probability of a home win is p = 0.68 and the offered odds are 1.35.
- 1Implied probability = 1 / 1.35 = 0.74074, or 74.074%.
- 2Edge in percentage points = 100 x (0.68 - 0.74074) = -6.074 pp.
- 3EV = 0.68 x 1.35 - 1 = 0.918 - 1 = -0.082, or -8.2%.
- 4Full Kelly = max(0, (0.68 x 1.35 - 1) / (1.35 - 1)) = max(0, -0.082 / 0.35) = 0.
Conclusion: although 68% is a high chance, odds of 1.35 overprice it. A high probability alone does not mean value; what matters is its relation to the price. The break-even odds for this estimate are 1 / 0.68 = 1.47. The same logic drives "play from": 1 / blended probability. This arithmetic applies to binary markets; do not generalise it to push or split markets. You can try your own numbers in the calculator on /learn.
How Slyx filters and what it measures
A pick passes only when expected value is at least 1% with a sharp book present and at least 3% without one; picks with non-positive EV or an AI veto are dropped. The limits are at most 15 analyses per day, 5 per league and one pick per event. When fewer picks qualify we publish fewer, or none; the limit is never filled artificially.
Results are public. Over the last 30 days there were 143 settled picks, a hit rate of 53.8%, a flat-stake ROI of 10.1% and an average CLV of 0.9% across 76 picks with a closing price. Over the last 90 days, though, it is 277 settled picks, a hit rate of 50.2%, an ROI of -1.5% and an average CLV of 1.1% across 145 picks. The shorter window looks better, the longer one is slightly negative, and that is exactly why neither can be sold as proof of skill. Slyx has not demonstrated independent profitability. Every loss stays visible on /historie, and the weekly bankroll reset does not erase them. The full overview lives in the results hub.
What this does not tell you: limits of verification
- Samples are small. For MLS there are only 9 settled picks in both the last 30 and last 90 days, so the percentages say little.
- A hit rate without odds is not enough. A 50% hit rate can lose or win money depending on average odds.
- CLV is a price-quality indicator, not a guarantee of results.
- The bankroll at /bankroll is a model portfolio of flat 500 CZK tickets per analysis, not proof of money actually wagered.
- Probabilities are estimates. A figure like 0.68 looks exact, but it is not.
We store predictions before kickoff and evaluate them later on newer data (Brier score, log loss). Brier = (p - outcome)^2, so a pick with p = 0.68 that comes in scores (0.68 - 1)^2 = 0.1024. A single game says nothing about calibration, though; only hundreds of predictions do. No model is switched automatically without review.
A checklist before you trust an analysis
- 1Find the source of each claim and the date it was published.
- 2Compute implied probability as 1 / odds and compare it with the estimate.
- 3Check whether the price is margin-free and how many bookmakers it rests on.
- 4Check timing: when the odds were captured and whether they have moved since.
- 5Compute EV rather than relying on probability alone.
- 6Look at the full record including losses, and at the sample size.
- 7Set a limit you will not exceed; see responsible betting.
The methodology page explains evaluation in more detail, and /jak-to-funguje gives the five-step overview. Slyx is for people aged 18 and over, and nothing here promises a return.
Reader questions
Is a high probability enough for a pick to be good?
No. What matters is the probability relative to the odds. At p = 0.68 and odds of 1.35, expected value is negative (-8.2%, illustrative example) because the odds demand a probability above 74.1%. A good pick has positive EV against the fair price, not just a high chance.
Why does Slyx show the odds at publication?
Because odds move, and an analysis can only be judged against the price the reader saw. Later we compare that price with the closing odds, which gives CLV. The average is on the track-record page; the per-pick chart is premium.
Do the last 30 days prove that Slyx works?
No. Over the last 30 days ROI is 10.1% across 143 picks, but over the last 90 days it is -1.5% across 277 picks. These samples are short and noise is large. Slyx has not demonstrated independent profitability and guarantees nothing.
Sources and methodology
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.