Picks powered by AI intelligence
How a release earns a place on the card.
Lock City Bet does not ask a model to feel confident. It asks five research agents to do assigned reading, write a fair line, and then survive a price check. The free pick of the day is the one release each morning that clears the public bar. The rest of the card is sold one pick at a time, or by subscription.
01 · Before 5 a.m. ET
Ingest
Schedules, injury reports, weather, umpire assignments, and opening lines land in the file. Nothing is a pick yet.
02 · Agents work alone
Independent scores
Each agent prices its own book without seeing the others. VEGA's football number does not borrow confidence from LYRA.
03 · Lock City Bet's price
Fair line
A release starts from Lock City Bet's number. If the board cannot write a fair line, the game is a pass.
04 · ORION signs or stops it
Market check
ORION compares the fair line with the live price. A moved number with no edge left is a pass, however pretty the matchup.
05 · Two agents, or it waits
Release rule
A side, total, or prop is published only when two agents agree and the price still clears. One clever model is not a release.
06 · 9 a.m. ET
One free pick
The highest-conviction release that also clears the public bar is the free pick of the day. The rest of the card is sold one pick at a time, or by the month.
The edges that count
NFL side
At least 1.0 point between the fair line and the posted number.
NFL total
At least 1.5 points, with the weather file already in.
NBA side
At least 1.5 points, and a rest or lineup assumption LYRA will stand behind.
Soccer total
At least 0.3 expected goals between JUNO's range and the posted total.
Player prop
The projection clears the number, and the minutes assumption is stable.
Unit size
1.0 is standard. 1.5 only when the edge is wide and ORION's price check is clean. Nothing above 1.5.
What the research count means
A source file might be a five-minute line history, a snap chart, a weather hour, a pitch log, or an expected-goals split. A typical release reads between 4,000 and 15,000 of them, scores a few hundred signals, and spends two to six hours of model time. The free pick on the Sunday card is at the high end of that range because football plus a market replay is a long file.
The count is published so you can see the work. It is not a charm. A 12,000-file pick can lose to a divisional game that turns into a punt fest. The record page keeps those losses next to the wins.
Passes are part of the method and they are not graded as wins. An agent that refuses a card is doing the job. JUNO will not freelance outside its file. CASTOR will not price a total before the starters are confirmed. LYRA will not publish a prop with a shaky minute band.