How We Grade Every Signal: Our Radical Transparency Approach

Every signal graded against final scores. Wins and losses, publicly tracked. Here's exactly how it works.

Most betting analytics platforms publish picks. Very few grade them. Even fewer publish both wins and losses publicly, in real time, with no ability to retroactively remove losing picks.

We do all of it. Every signal that reaches the dashboard gets graded against the final score. Every result - win, loss, or push - is tracked on our Track Recordpage. There's no editing, no deletion, no cherry-picking. Here's how the system works.

The Grading Pipeline

Our system runs multiple times daily. When a game ends, the grading process kicks off within the next cycle:

Step 1: Score Collection (Three Sources)

We learned early that relying on a single API for final scores creates grading bottlenecks. If one source goes down or has delayed updates, signals sit ungraded for days. So we built a three-source scoring system:

If the primary source has the score, we use it. If not, we check the secondary, then tertiary. This triple-redundancy means signals rarely sit ungraded for more than a few hours after the game ends.

Step 2: Signal Matching

Each signal is linked to a specific game by game ID and matchup string. The grading system uses fuzzy team name matching to handle naming inconsistencies between data providers (for example, "Conn." vs "Connecticut" vs "UConn"). Once matched, the signal's pick is compared against the final score.

Step 3: Result Determination

The grading logic is straightforward and transparent:

Step 4: Unit Calculation

Every pick is to win 1 unit at standard -110 juice. Win +1.0u, lose -1.1u. No variable sizing, no confidence-based multipliers. This keeps the math clean and performance comparisons apples-to-apples. Simple, standardized, fully transparent.

What We Track (And Show You)

Every graded signal contributes to our public performance metrics:

How Grading Makes Our Models Sharper

This is where it gets interesting. Grading isn't just for the scoreboard - it's the input that makes our models improve automatically.

After every grading cycle, our system recalibrates the weights for each model type. Here's the simplified logic:

This creates a self-correcting system. Signal types that work keep producing. Signal types that don't get automatically demoted. No manual intervention required - though our admin panel allows manual overrides when we identify specific market conditions that explain temporary underperformance.

Why This Matters

The sports betting analytics market is full of platforms that claim impressive win rates without showing their work. We've seen competitors claim "80% accuracy" without publishing a verifiable track record. Others only show their best days and quietly remove losing streaks.

Our approach is different because it has to be. Every published pick was reviewed by our analysts before tip-off - and graded automatically against final scores after. No retroactive editing. No cherry-picking. What you see on the Track Record page is the complete record - picks our analysts called, scored by the math.

We believe this is the right way to build trust. Not through marketing claims, but through verifiable results - both the wins and the losses - tracked in real time. That's what "decision support, not picks" really means: giving you the data to evaluate our models yourself, rather than asking you to take our word for it.