HOW THE SAUSAGE GETS MADE · no gut feel · no hot hands
We bet numbers,
not teams.
Every recommendation walks the same path: validate the market row, compare probability with price, require a positive model stake, and preserve the recorded outcome. Here's the pipeline and the distinction between Board recommendations and the broader benchmark.
Build a complete priced signal
A recommendation starts only after the player, game, prop, side, line, and real book price resolve to one verified market row.
The projection and market services join on stable player and game identities. Incomplete or unpriced rows remain research data; they cannot become public recommendations.
The same captured row later receives the closing price and recorded box score, which lets the Market Memory and Ledger compare line versus actual without changing identities after the game.
Price decides whether it is actionable
Hit likelihood alone is not value. The model compares its calibrated probability with the exact price's breakeven rate.
A heavy favorite can be likely to hit and still be badly priced. A plus-money prop can sit below 52% and still offer value. That is why the likelihood band and the price-relative value edge stay separate.
Prices below the service's minimum payout are excluded. For the remaining candidates, the grading service computes a fractional-Kelly stake. Zero is a real no-bet answer, not a missing value to replace.
Publish the positive-stake subset
The Board shows only complete CORE rows with a positive recommended stake, deduplicated to one quote per physical player-market.
Alternate lines for the same player and market compete with each other; they do not become duplicate recommendations. The slate cap is applied only after every scheduled game has had a chance to contribute.
The Ledger answers a different question: how every eligible priced signal from the currently served CORE formula performed at a normalized flat 1u. Naming both populations prevents a model benchmark from being mistaken for the smaller recommended card.
HOW WE GRADE · the part most touts skip
A pick isn't real
until it's settled.
Recommendations are time-stamped when posted and settled from the recorded box score after the game. Closing-line value is attached when that feed is available. Here's the same rulebook whether the play wins or loses.
PLAIN ENGLISH
The four words
that matter.
One normal bet for your bankroll — many use 1% of it. Sizing in units means our record reads the same whether you bet $5 or $500. We never tell you a dollar amount.
The calibrated probability that the selected side wins. It is not an edge score: price determines whether that probability is worth betting.
The difference between model hit probability and the exact price's breakeven probability. A positive value can earn a model-sized stake; a zero or negative value stays off the recommendation Board.
Every eligible priced signal from the formula serving CORE today, graded at a normalized flat 1u. It is broader than the positively staked recommendations shown on the Board.
See the annotated design example
This archived NBA Finals example explains the intended anatomy of a detailed pick receipt. It is a labeled product walkthrough, not a row from the current league's live Board or Ledger, and opening it does not change your selected league.
Read a worked example →STILL SKEPTICAL? · good
Don't trust us.
Read the ledger.
The method only matters if it shows up in the results. Read the current CORE formula's full eligible-signal benchmark, then compare it with the smaller actionable Board.