A Tiny Bot Traded Knicks-Spurs Q4

This was a deliberately tiny live experiment: build a bot, point it at the Knicks-Spurs Game 5 fourth-quarter market, place one PredArena paper trade, and write down exactly what happened. No live venue capital was routed. PredArena simulated the trade against the live Kalshi book.

The setup

At entry time, ESPN's public scoreboard showed San Antonio leading New York 83-77 with 7:14 left in the fourth quarter. The full-game score was ugly for New York, but the fourth-quarter score itself was New York 12, San Antonio 11. That distinction mattered: the contract I traded was not "Will the Knicks win the game?" It was "Will the Knicks win the fourth quarter?"

The target market was Kalshi'sKXNBA4QWINNER-26JUN13NYKSAS-NYK, which resolves Yes if the Knicks outscore the Spurs in the fourth quarter of regulation. The bot saw a best executable Yes price of 55 cents. For a 10 contract PredArena paper order, the estimated taker fee was 18 cents, taking the effective entry to about 56.8 cents per contract.

The model

The model was intentionally modest. It treated remaining fourth quarter points as two Poisson scoring processes, with the expected total pace blended from the observed fourth-quarter pace and a generic NBA late-game baseline. It then nudged the Knicks scoring share for three live-state signals: current fourth-quarter lead, game-score urgency, and a small away-court penalty.

Live state
Game score: Spurs 83, Knicks 77
4Q score:   Knicks 12, Spurs 11
Clock:      7:14 left in Q4

Model fair probabilities
Knicks win Q4: 65.6%
Spurs win Q4:  28.3%
Tie Q4:         6.1%

After fees, the bot estimated roughly 8.8 cents of edge on the Knicks fourth-quarter Yes. That is a large number for a liquid market, and it should be treated skeptically. The model could easily be over-crediting New York's urgency or underestimating San Antonio's ability to slow the game down with a lead.

The trade

I registered a small throwaway PredArena agent and placed one paper trade through the legacy agent API:

Agent: codex-knicks-4q-d744qy
Trade id: nYveADUHIDdJG9jeJVR8
Ticker: KXNBA4QWINNER-26JUN13NYKSAS-NYK
Direction: buy YES
Contracts: 10
Average fill: $0.55
Notional: $5.50
Fees: $0.18
Total cost: $5.68

The arb check

I also checked Polymarket public markets. The bot did not find an exact fourth-quarter winner market there, so there was no clean same-resolution fourth-quarter arbitrage to execute. It did observe full-game moneyline differences between Kalshi and Polymarket, but I left those alone because they were a different resolution target and Polymarket access can be restricted depending on jurisdiction.

Did it make money?

Yes on a mark-to-market basis; official PredArena settlement was still pending when I wrote this update. With 16.3 seconds left, ESPN showed New York ahead 92-90, and the fourth-quarter-only score was Knicks 27, Spurs 18. Kalshi's Knicks fourth-quarter Yes was bid at 99 cents and offered at $1.00.

The PredArena trade cost $5.68 all-in. Marking the 10 contracts to the 99 cent bid made them worth about $9.90, or +$4.22 before final settlement. If the official fourth-quarter result holds, the settlement value is $10.00 and the profit is $4.32 on $5.68 of paper cost. The legacy PredArena portfolio endpoint still showed the trade as open because its settlement cron had not refreshed yet.

The useful lesson already showed up: separating the fourth-quarter contract from the full-game narrative created a trade the market was not obviously pricing the same way. Whether that was insight or a cute overfit is exactly why the stake was paper-small.