Data hold
NFL
Awaiting licensed feed
- Engine
- Drive-level football simulation
- OOS log loss
- No champion yet
Sports prediction intelligence · not a sportsbook
Probabilities
you can audit.
Every game simulated 250,000 times by a sport-native engine, calibrated on seasons the model never saw, and written to a public hash-chained ledger before it starts. Wins, losses and holds stay on the record.
Live status is reconnecting…
Next up
Today’s board
The next games across every sport with a live model. Locked cards show exactly what a higher tier unlocks — every pick becomes public at game time.
The slate is temporarily unavailable. The intelligence service is not responding. This page will show live data as soon as it reconnects.
Coverage
Nine sports. True states.
A sport goes live only when a champion model has passed its promotion gates and games are on the schedule. Everything else says exactly why it is waiting.
Data hold
College Football
Awaiting licensed feed
- Engine
- Drive-level football simulation (college rules)
- OOS log loss
- No champion yet
Data hold
NBA
Awaiting licensed feed
- Engine
- Possession-level basketball simulation
- OOS log loss
- No champion yet
Data hold
College Basketball
Awaiting licensed feed
- Engine
- Possession-level basketball simulation (college rules)
- OOS log loss
- No champion yet
Data hold
MLB
Awaiting licensed feed
- Engine
- Half-inning baseball simulation
- OOS log loss
- No champion yet
Data hold
NHL
Awaiting licensed feed
- Engine
- Period-level hockey simulation
- OOS log loss
- No champion yet
Data hold
Soccer
Awaiting licensed feed
- Engine
- Dixon–Coles goal model
- OOS log loss
- No champion yet
Data hold
Tennis
Awaiting licensed feed
- Engine
- Point-level Markov tennis model
- OOS log loss
- No champion yet
Data hold
Horse Racing
Awaiting licensed feed
- Engine
- Correlated field simulation with pace scenarios
- OOS log loss
- No champion yet
The protocol
How the engine works
Eight stages, the same for every sport. Each one can stop a pick; none of them can quietly change one after it is published.
-
01
Point-in-time data
Every input is stamped with when it was known. Training and prediction only see what existed before the event.
-
02
Sport-native state
Uncertainty-aware team ratings plus measured context: rest, travel, altitude, availability.
-
03
250,000-trial simulation
Drives, possessions, half-innings, periods, goals or points — played out with a recorded seed.
-
04
Calibration
Raw probabilities are recalibrated on walk-forward seasons so 60% means 60%.
-
05
Adversarial review
Stress scenarios try to flip the pick: a key absence, ratings one sigma wrong, a low-event game.
-
06
Immutable ledger
Published or held, every decision is hashed and chained before the event starts.
-
07
Grading
Finals are graded against the stored pick, never a revised one. Corrections are new rows.
-
08
Learning
Ratings update after each result; candidates must beat the champion on held-out data to be promoted.
Out-of-sample
Performance snapshot
Walk-forward backtests: each season is predicted by a model trained only on earlier seasons. Lower log loss is better; the baseline always picks the historical home-win rate.
Performance data is temporarily unavailable. The intelligence service is not responding. This page will show live data as soon as it reconnects.
Receipts
A ledger that cannot be quietly edited
Each prediction row stores a SHA-256 hash of its own contents plus the hash of the row before it. Change one character in any past pick and every hash after it stops matching. The database also refuses updates and deletes on the ledger tables outright — corrections are new rows that point at what they correct.
- Written before the event starts — the database rejects late pregame rows
- Data holds are recorded too, with their reasons
- Every locked pick becomes public at game time
The ledger is empty.
Membership
Start free. Go deeper when it pays.
The ledger, results and calibration are free forever. Paid tiers unlock the full model output before kickoff.
Questions
Straight answers
Is VigLines a sportsbook?
No. VigLines publishes probabilities and analysis. We do not accept wagers, hold funds or take a side in any bet.
How do I know picks were not changed after the game?
Every prediction is written to an append-only ledger before the event starts. Each row stores the SHA-256 hash of its contents and the hash of the row before it, so editing any past row breaks every hash after it. Anyone can run the verification on our ledger page.
Why are some sports on hold?
We only publish when the model has passed its promotion gates and the live data is licensed for the use. When either condition fails, the sport or the game is shown as a data hold with the reason, instead of a guess.
What does a 60% probability mean?
Across many games where the model says 60%, that side should win about six times in ten. Our calibration charts show how closely that has held in out-of-sample testing.