College Football

Method. Three free feeds, none of them paid for and none of them interchangeable:

• Action Network — consensus % of money and % of tickets, plus a timestamped history of both the number and the price, at seven books individually rather than only the consensus.
• DraftKings — DraftKings' own handle % and bet %, independent of the consensus above, which is what makes agreement between them worth anything.
• Pinnacle — the sharp reference price. It takes sharp action and moves on it instead of shading to balance a public book's exposure, so it answers a different question: not where the money is, but where the number should be.

What gets flagged. A game needs at least 8,000 tracked bets — on thin markets the two splits feeds disagree by a median of roughly 30 points of money share, so their percentages there cannot be trusted — and both feeds must agree on which side holds the money, with at least 60% of the money on it. At 51/49 a book holds no meaningful exposure either way: it is balanced and collecting vig from both sides, so moving the number “against the money” there is not a statement, it is noise. Then either the number moved against that money in the last 3 days, or the price did by at least half a point's worth: a book shades the vig before it moves the line, and a spread sitting still at −3 while the price runs −110 to −125 has moved. What scores it. The move is weighted by the key numbers it touched, because giving up 3 is not the same as drifting through 10. It is then scaled by how lopsided the money is, how much of it there is, whether the play is getting a better number than Pinnacle, and whether several books moved together within half an hour — the market being hit in parallel rather than one book drifting. That produces the 0–10 Conviction score. 2.5 and above is an official pick; everything below is listed and graded but kept out of the headline record.