Probabilistic forecasting engine · 2024

Predict with an edge.

Wagr turns historical patterns and live market signals into calibrated odds — then lets you and your friends make play-money predictions on real-world events, global markets, and each other. No cash, no bookie. Just a sharper read on what happens next.

Play-money only Real-time odds Built with PyTorch
PythonPyTorchPerplexity API Real-time dataCalibrated probabilities
9:41
0
markets priced live
3.1%
expected calibration error
0
historical + live features
v4.2
PyTorch model release
How it works

From a hunch to a calibrated wager in three taps

Spin up a market on any event, let the model price the odds, and settle up with your friends — all in play-money coins.

01

Create or join a market

Open a peer-to-peer challenge with a friend, or join a global market where the whole crowd is predicting the same event.

02

The model prices the odds

A PyTorch network fuses historical form with live market telemetry and posts a calibrated win probability — alongside an AI news briefing for context.

03

Predict & settle with friends

Stake play-money coins where you spot an edge, track open predictions in your portfolio, and climb the friends leaderboard.

The model

A forecasting pipeline you can actually read

Most prediction engines are a black box. Wagr keeps historical features, live signals, and the generated probability in separate, inspectable stages — so every number on screen is traceable back to its inputs.

HISTORICAL FEATURES trained weights
Elo & recent form Head-to-head, 5y Home / away splits Rest & schedule Lineup strength Pace & style fit
LIVE SIGNALS real-time
Bookmaker line moves Market volume Crowd order flow Conditions / weather Late lineup news Sentiment shift
MODEL OUTPUT PyTorch · v4.2
0%P(WIN)
Synthesizing historical features with live signals, the network outputs a calibrated win probability with an 80% prediction interval — not just a point guess.

Interpretable by design

Separated stages. Historical inputs, live signals, and the generated probability are computed and surfaced independently, making each contribution easy to audit.
Custom PyTorch network. A purpose-built model learns from years of event data and blends it with external market signals into a single probability estimate.
Calibrated, not just confident. Outputs are probability-calibrated and bounded by an 80% prediction interval, so a posted 63% really means 63%.
Edge, made obvious. The model's fair odds sit next to the crowd's price, so the gap — your edge — is visible before you ever commit a coin.
Calibration is tracked continuously: across resolved markets, predicted probabilities land within a few points of observed frequencies.
Features

A prediction game built for your group chat

Predict on real-world events, on global markets, or on each other — with a model pricing the odds, gamified progress, and AI context for every pick.

Friend props

Predict on your crew's real life — "Marcus runs a sub-22 5K by Sunday," "Dana ships her side project." Set the line and let the group price it.

Friend duels

Call out a friend head-to-head. The model shows both sides a fair line, then it's you vs. them — with bragging rights and a win-loss record on the line.

Parlay builder

Stack multiple predictions into one slip. Wagr multiplies the legs into a combined probability — smaller chance, bigger play-money payout.

Global markets & order book

Join crowd-wide markets backed by a live order book — crowd vs. model fair price, recent predictions, and depth, all in real time.

AI news briefings

Before you predict, an automated summary — generated with the Perplexity API — distills the latest news on the event into a few sharp bullets.

Streaks, XP & quests

Daily quests, win streaks, and XP that levels you up. It's built like a game — because it's for fun and bragging rights, not for cash.

Portfolio & P&L

Track open predictions, settled history, and all-time play-money profit with a clean equity curve — sortable by the model's edge.

Friends & leaderboard

Add friends, duel them directly, and see who's actually sharp this season on a ranked, season-long leaderboard with levels and streaks.

Live odds & edges

Odds update as signals move. The home feed surfaces the model's top edges of the day so you never miss a mispriced market.

Play, don't gamble

Built like a game, not a sportsbook

Wagr runs entirely on play-money coins. Progress comes from streaks, levels, and bragging rights with your friends — there's nothing to deposit and nothing to lose but a duel.

14-day
current win streak
Level 12
Sharpshooter tier
3 / 3
daily quests done
AI news briefings

Walk in with the context, not just the odds

For every major market, Wagr calls the Perplexity API to generate a concise, sourced briefing of what's happening right now — injuries, momentum, conditions — so the number on screen comes with a story. It's the difference between guessing and knowing why the model moved.

Auto-generated and refreshed as the event approaches
Distilled to the few facts that actually move a probability
PERPLEXITY · AI BRIEFING 3 sources · updated 4m ago
Lions FC vs. Harbor United — Match winner
Lions FC enter on a four-match unbeaten run, scoring first in each of their last five at home.
Harbor United will be without their first-choice keeper (suspension) — backup has conceded in every away start.
Forecast rain is expected to slow the pace, historically favoring the stronger set-piece side.
Crowd consensus has firmed on Lions over the last 24 hours — from roughly 54% to 58% implied.
Synthesized for Wagr AI▲ parsed into live-signal flags
Interactive demo

Take the app for a spin

A working slice of the Wagr mobile app — tap through the screens and watch the live odds move. All data is illustrative — play-money only, no real currency.

Jump to a screen
9:41
Under the hood

The stack behind the forecast

A real-time pipeline that turns raw event data and market telemetry into calibrated, interpretable probabilities.

SOURCESHistorical events + bookmaker feeds
INGESTReal-time data pipeline
MODELPyTorch network
APPCalibrated odds + briefings
Modeling
Custom PyTorch NN
A purpose-built network that synthesizes historical patterns and external market signals into event probability estimates.
Data
Real-time pipeline
Combines historical event data with live bookmaker telemetry, continuously refreshing inputs as signals move.
Context
Perplexity API
Automated AI-generated news summaries for major events. A lightweight parsing step turns each briefing into structured flags — injury, suspension, weather — that join the live-signal layer.
Markets
P2P + global flows
Market creation, odds updates, and order-book settlement powering both head-to-head and crowd-wide predictions.
Model live · 1,284 markets priced

See where the model finds an edge

Tap through the interactive demo to watch historical patterns and live signals collapse into a single calibrated number.