The Engine
Project 262
An ensemble system built to find value where the market misprices it.
Overview
What is Project 262
Project 262 is an ensemble machine learning system designed for sports markets — currently focused on the NHL, with additional leagues planned.
It doesn't rely on one model or a “magic formula.” It runs 262 independent predictive filters, each evaluating the game from a different angle. Before each slate, the filters cast their signals and the system aggregates them into a final portfolio.
Consensus Strength
More mandates → stronger consensus → larger allocation
Risk Management
Ensemble Hedging
Sometimes the system allocates units to both sides in the same match.
This isn't a glitch — it's variance control. Different filters can detect value in opposite directions of the market line. Instead of discarding one side, the engine distributes units to capture expected value while managing risk.
NJ Devils vs CBJ Blue Jackets · Feb 4, 2026
NJ Devils
ODDS: 1.72
VOTES: 1
4.22u
CBJ Blue Jackets
ODDS: 2.02
VOTES: 1
10.78u
Both sides receive allocation — the system captures value from the spread
Allocation
Kelly Lite
We don't use flat sizing. Every signal comes with a precise unit allocation computed by a conservative Kelly-based model.
The daily budget is 100 units. The system distributes them across the day's portfolio based on consensus strength and price value.
Golden rule: follow the portfolio. Cherry-picking breaks the math.
DAILY BUDGET
100 units
Transparency Boundary
What We Share — and What We Don't
✦ Public
- Signal direction (which side)
- Consensus strength (mandates)
- Exact unit allocation
- Market odds at signal time
- Full performance history
- Blockchain verification files
◆ Private
- Individual filter logic
- Training data and features
- Model architecture details
- Implementation internals
The edge lives in the implementation, and we keep it there.
See the system in action
Daily signals, real results, and public verification.
Join Discord