How BAI8 prices a fixture
A fitted model produces a number, deep research produces a second one, and the two are composed rather than averaged. The whole pipeline, in order.
Every assessment this product publishes is the same four steps in the same order. Nothing about it is secret, and the reason to write it down is that a number you cannot interrogate is a number you have to take on faith.
One: the fitted model
The model is fitted on results, and only on results. For a football match it estimates each side's attacking and defensive rate, adjusts for where the game is played, and runs the two forward into a distribution over scorelines. Every market in the fixture is then read off that same distribution, which is what keeps the match odds, the totals and the handicaps consistent with each other.
This half is fast, cheap and completely blind. It does not know that a manager was sacked on Tuesday. It knows what teams like these have done against teams like those, and that is a genuinely strong starting point and a hard ceiling.
Two: the research passes
Then thirteen agents read everything published about the fixture: team news, press conferences, local reporting in the local language, injury records, travel, and the fixture congestion either side of it. Each pass returns claims with sources attached, and a claim without a source is discarded rather than softened.
Three: composition
The two numbers are combined, weighted by how much the research actually found. A fixture where nothing was learned stays close to the model. A fixture where three sources agree that a first-choice defence is unavailable moves, and the size of the move is a function of the evidence rather than of how interesting the story was.
Four: the price
Only at the end does the market enter. The composed probability is compared to what your book is charging, and the difference is the edge, which is the gap between two numbers and nothing more than that.
What this does not tell you
The pipeline is not a guarantee of accuracy at any step. A model fitted on thin data is confidently wrong, research can find three sources repeating one rumour, and a composition of two bad estimates is a bad estimate. What the order buys is that every figure has a stated provenance, so when it is wrong it is possible to say which step was wrong.
Common questions
Does BAI8 use a model or research?
Both, and they are composed rather than averaged. The fitted model produces a base rate from history, research adjusts it for what history cannot see, and the composition is weighted by how much the research actually found.
Why not just use the model?
Because a model fitted on results has no way of knowing that a first-choice goalkeeper was ruled out this morning. Research is how information that has not yet reached the data gets into the number.
Does BAI8 place bets?
No. It publishes a number and the reasoning behind it, and holds no bookmaker account. Nothing it produces is an instruction.