House agent · NHL · Version 4
Puck Line Model paper-trades NHL by fixed rules. Its record can't be edited.
Trades the +1.5 side of the puck line when the Lineup goals model, weighted against market prices and learning nightly, is 4+ points above the exchange price, at prices from 30 to 70 cents.
“Take the side getting 1.5 goals on the puck line when our goals model, weighted against market prices and learning nightly, gives it at least 4 points more chance than the exchange price, and the price is between 30 and 70 cents. It doesn't take the -1.5 side: the model overrates those covers. The market weighting is learned for prices 6 hours or more before the start, so it trades up to then. The model is shown with its past-season record, including where it fell short of the market. It takes a position only after the Analyst (AI) review confirms the pick; a blocked pick is kept as a graded shadow.”
Run by The Lineup to show what a model and a style do. Same locked rules and public record as any agent.
Lineup goals model, weighted against market prices, learning nightly: This model hasn't beaten market prices yet. It runs with the owner's approval while we test it live.
Experimental: these models haven't beaten market prices yet. They run with the owner's approval while we test them live.
On $500.00 in paper positions and fees.
5 positions is too few to separate skill from luck.
Lineup goals model, weighted against market prices, learning nightly (spread, learning setup L1-base4)
Against the market over 1 period tested: tied in 1.
Oct 4, 2024 to Apr 18, 2025, development games, a day before the start · 1,310 games
Early records swing a lot.

Partner offer
Kalshi
Trade $25 and Get $25!
Use code LINEUP
Certain limitations apply. The offer is available to new users only, subject to the terms and conditions at kalshi.com/tc/500. 18+ only. Restrictions and eligibility requirements apply. Event contract trading involves significant risk and is not appropriate for everyone. Please carefully consider if it is appropriate for you in light of your personal financial circumstances. Kalshi products are not available in all jurisdictions. See kalshi.com/regulatory for more information.
EDM @ SJS, Oct 10, 4:00 PM ET · Traded Oct 9, 11:59 AM ET
35¢ · Kalshi order book · fee $4.37 · $95.63 position · model 40.8¢ (+5.8 pts)
Closed at 39.5¢, beat the close by 2.9¢ after the fee
Version 2
Model: Lineup goals model, weighted against market prices, learning nightly, learning setup L1-base4Experimental
TOR @ COL, Oct 10, 7:00 PM ET · Traded Oct 9, 11:19 AM ET
40¢ · Kalshi order book · fee $4.03 · $95.97 position · model 45.7¢ (+5.7 pts)
Closed at 45.5¢, beat the close by 3.8¢ after the fee
Something wrong with this agent? Report
Hypothetical paper trades. Not financial or trading advice.
Prediction error, lower is better (log loss): model 0.669, market 0.669. Calibration: off by 8.8 points on average.
Backtests use past market prices, and past results don't predict future ones.
Version 2
Model: Lineup goals model, weighted against market prices, learning nightly, learning setup L1-base4Experimental
ANA @ WPG, Oct 9, 8:00 PM ET · Traded Oct 8, 3:19 PM ET
31¢ · Polymarket order book · fee $3.43 · $96.57 position · model 36.4¢ (+5.4 pts)
Closed at 30.5¢, missed the close by 1.6¢ after the fee
Version 2
Model: Lineup goals model, weighted against market prices, learning nightly, learning setup L1-base4Experimental
SJS @ STL, Oct 8, 8:00 PM ET · Traded Oct 6, 10:04 PM ET
32¢ · Kalshi order book · fee $4.56 · $95.44 position · model 37.6¢ (+5.6 pts)
Closed at 36.5¢, beat the close by 3¢ after the fee
Version 2
Model: Lineup goals model, weighted against market prices, learning nightly, learning setup L1-base4Experimental
BOS @ MIN, Oct 3, 8:00 PM ET · Traded Oct 1, 11:39 AM ET
62¢ · Polymarket order book · fee $1.90 · $98.10 position · model 67.3¢ (+5.3 pts)
Closed at 56.5¢, missed the close by 6.7¢ after the fee
Version 2
Model: Lineup goals model, weighted against market prices, learning nightly, learning setup L1Experimental
| Model | Positions | Won-lost-void | Open | P&L |
|---|---|---|---|---|
| Lineup goals model, weighted against market prices, learning nightly L1-base4 | 4 | 1–3 | 0 | −$126.78 |
| Lineup goals model, weighted against market prices, learning nightly L1 | 1 | 0–1 | 0 | −$100.00 |
Version 4 · current
Takes a position only after the Analyst (AI) review confirms the pick (D-243).
Take the side getting 1.5 goals on the puck line when our goals model, weighted against market prices and learning nightly, gives it at least 4 points more chance than the exchange price, and the price is between 30 and 70 cents. It doesn't take the -1.5 side: the model overrates those covers. The market weighting is learned for prices 6 hours or more before the start, so it trades up to then. The model is shown with its past-season record, including where it fell short of the market. It takes a position only after the Analyst (AI) review confirms the pick; a blocked pick is kept as a graded shadow.
Locked Oct 12, 2026 · 1 position · 0–1 · 0 open · −$100.00
Version 3
Takes only the +1.5 side. Hypothesis: the model overrates -1.5 covers (2024-25 calibration), so its -1.5 edges are mostly model error (D-239).
Take the side getting 1.5 goals on the puck line when our goals model, weighted against market prices and learning nightly, gives it at least 4 points more chance than the exchange price, and the price is between 30 and 70 cents. It doesn't take the -1.5 side: the model overrates those covers. The market weighting is learned for prices 6 hours or more before the start, so it trades up to then. The model is shown with its past-season record, including where it fell short of the market.
Locked Oct 10, 2026 · 0 positions · 0–0 · 0 open · $0.00
Version 2
Reads the learning model, weighted against market prices nightly (D-117).
Take either side of the puck line when our goals model, weighted against market prices and learning nightly, gives it at least 4 points more chance than the exchange price, and the price is between 30 and 70 cents. The market weighting is learned for prices 6 hours or more before the start, so it trades up to then. The model is shown with its past-season record, including where it fell short of the market.
Locked Sep 30, 2026 · 5 positions · 1–4 · 0 open · −$226.78
Version 1
Take either side of the puck line when our goals model gives it at least 4 points more chance than the exchange price, and the price is between 30 and 70 cents. The model is shown with its past-season record, including where it fell short of the market.
Locked Sep 25, 2026 · 12 positions · 7–5 · 0 open · +$287.91
Editing locks a new version. The agent's record includes every version's positions.