Methodology
How a match is actually analysed
Four steps, in order. The interesting part is not the model — it is what happens to the numbers before the model sees them.
- 01
Collect the whole market, not one price
Sharkline reads the odds several dozen bookmakers publish for the same match, rather than the price of a single firm.
One bookmaker’s price is one company’s opinion plus its own risk position. The aggregate of many is closer to what the market as a whole believes. Where books disagree sharply, that disagreement is itself a signal worth reporting.
- 02
Strip the built-in margin
Published odds always contain the bookmaker’s cut, so the implied probabilities add up to more than 100%. That excess is removed mathematically before anything else happens.
The industry calls the cut the vig or the overround. On a two-way market the raw implied probabilities might total 105%; the extra five points are the house edge, not information about the game. Removing them leaves probabilities that can honestly be compared with a model’s own estimate.
- 03
Analyse on cleaned data plus match context
The model works from the de-margined probabilities together with squad, form, head-to-head and competition context.
It is explicitly constrained to the numbers present in the sources. When a source has no data for a fixture — a minor competition, an obscure market — the system is required to say so rather than fill the gap with a plausible-sounding invention.
- 04
State a confidence level and a risk factor
Every analysis carries one of four confidence levels and names the single factor most likely to make it wrong.
The levels are Strong, Moderate, Speculative and Avoid. Avoid is a real outcome that the system reaches regularly, which is the part that separates an analytics tool from a tipster: a tipster is paid to always have an answer.
What it cannot do
The limits, stated plainly
- No result is guaranteed. A well-calibrated 70% call is still wrong three times in ten, by design.
- Thin markets carry thin data. Lower-tier competitions have fewer bookmakers pricing them, so the consensus is weaker and the confidence level reflects that.
- The model does not know about last-minute news it has not been given — a late injury or a rotated squad announced after the analysis will not be in it.