Putting an Entire Season of Predictions to the Test
Throughout High Point’s historic season, every game preview in our Toothsayer series included a projected final score with a projected point spread and an outright winner.
The Toothsayers were always designed to be informative preview pieces. Final score projections served as a confidence meter—blending advanced analytics, matchup dynamics, and situational context—rather than explicit betting picks.
Still, with a full season in the books and one on the way, it’s time to grade our work. Both to give a perspective of how we did and also to see where and what can be improved.
Importantly, every Toothsayer preview was written and published before sportsbooks released their opening betting lines. These predictions weren’t made after Vegas established market consensus or after advanced metrics fully settled on a number. They were published independently, left on the record, and evaluated against the final outcome.
As we prepare for the upcoming season, we wanted to step back and answer a simple question: How accurate was the Toothsayer?
Rather than relying on selective memory, every game was measured against three objective benchmarks:
- Did the Toothsayer correctly predict the winner?
- How close were the projected margins to the final score?
- How did those projections compare to the closing Vegas betting line?
The goal isn’t to claim perfection. Every projection system has misses and mistakes. The goal is complete transparency; identifying where our process succeeded and where it can evolve.
2025–26 Toothsayer Results
| Metric | Result |
| Games Predicted | 26 |
| Straight-Up Record | 22–4 (84.6%) |
| Mean Absolute Error (Spread) | 12.52 points |
| Root Mean Squared Error | 15.22 points |
| Exact Margin Predictions | 1 |
| Predictions Within 3 Points | 4 |
| Games With Vegas Lines | 23 |
| Closer Than Closing Vegas Line | 12 of 23 (52.2%) |
| Timing of Publication | Before Vegas Opening Lines |
Picking the Winner
The simplest metric is often the least revealing, but it provides a necessary baseline.
Across 26 games, the Toothsayer correctly projected the winner 22 times, finishing with an 84.6% straight-up record.
The four prediction misses occurred in the following matchups:
- UAB (Projected HPU win; Actual: UAB win)
- Southern Illinois (Projected HPU win; Actual: SIU win)
- UNC Asheville (Projected UNCA win; Actual: HPU win)
- Longwood – Second Meeting (Projected Longwood win; Actual: HPU win)
Considering High Point spent much of the season ranked among the nation’s top mid-major programs and entered most conference games as a decisive favorite, a high straight-up win percentage was expected.
What mattered more was that the model didn’t blindly rubber-stamp Panther victories every night. Projections against UNC Asheville, Winthrop, and Longwood correctly identified spot-reversals and difficult situational matchups.
Whether those picks ultimately hit or missed, they proved the model was evaluating matchups objectively rather than simply reinforcing fan expectations.
Looking Beyond Wins and Losses
Predicting who wins is only part of the challenge. Predicting how a game unfolds is considerably more difficult and far more interesting for readers.
To evaluate calibration, every projected spread was compared against the actual final margin. The result was a Mean Absolute Error (MAE) of 12.52 points, meaning the average projection finished just over twelve points away from the eventual outcome.
While there is no universal benchmark for a “good” college basketball MAE, the closing betting market offers the most rigorous comparison available. Vegas closing lines posted an MAE of 11.26 points across the same sample—leaving the Toothsayer only 1.26 points farther from the final margin on an average night.
For independent projections published before sportsbooks set a market, that is a strong benchmark, even if bringing that error closer to 10 points remains our target for next season.
Some games landed nearly spot-on:
Closest Margin Predictions
| Opponent | Predicted Margin | Actual Margin | Error |
| USC Upstate | HPU +20 | HPU +20 | 0 |
| Winthrop | HPU +3 | HPU +2 | 1 |
| UIC | HPU +7 | HPU +10 | 3 |
| Radford | HPU +6 | HPU +9 | 3 |
On the other end of the spectrum, games like Bryant produced massive variance. High Point was projected to win by 13 before dismantling the Bulldogs by 46. Even the betting market severely underestimated that performance, setting the Panthers as only 19.5-point favorites.
Games like that illustrate a fundamental reality of college basketball analytics: once a superior team begins to separate, predicting the exact scale of a blowout becomes extraordinarily volatile.
The Benchmark That Matters Most
The most compelling comparison wasn’t against the final score, it was against the market.
Because Toothsayer projections were published before sportsbooks posted lines, they were generated completely free of market influence, public betting trends, or oddsmaker adjustments.
Among the 23 games with published closing spreads, the Toothsayer finished closer to the final margin 12 times, while the closing Vegas line finished closer 11 times.
| Comparison | Result |
| Toothsayer Closer to Final Margin | 12 (52.2%) |
| Closing Vegas Line Closer | 11 (47.8%) |
This shouldn’t be misconstrued as “beating Vegas.” A 23-game sample is far too small to draw sweeping conclusions about beating market efficiency long-term.
However, it does demonstrate that an independent projection model built before sportsbooks established a line can compete with the most efficient pricing mechanism in sports.
When you consider that closing lines integrate injury news, syndicate betting capital, and multi-million-dollar risk management, holding a 52.2% edge in head-to-head accuracy validates the Toothsayer process.
What Worked Well
Reviewing the data reveals several clear operational strengths:
- High-Accuracy in Competitive Matchups: The model performed best in games expected to be tight. Single-digit and low double-digit projections consistently produced our smallest error margins, often landing within a possession or two of the final horn.
- Macro Team Evaluation: The model successfully gauged High Point’s overall quality relative to its opponents throughout the entire campaign.
In short, the Toothsayer excelled at answering “Who is the better basketball team?” even when measuring the precise margin proved trickier.
Where the Model Can Improve
Every dataset presents an opportunity for refinement. The numbers highlight two clear areas for improvement:
1. Better Calibrating Blowout Ceilings
The clearest trend was that the Toothsayer systematically underestimated High Point’s ability to turn favorable matchups into utter routs. On average, our projected margin sat roughly 6.6 points lower than the actual result in runaway wins.
| Matchup | Predicted Margin | Actual Margin | Undervalued Margin |
| Bryant | HPU +13 | HPU +46 | +33 |
| Gardner-Webb | HPU +26 | HPU +55 | +29 |
| Averett | HPU +46 | HPU +75 | +29 |
| Canisius | HPU +18 | HPU +43 | +25 |
| Furman | HPU +5 | HPU +26 | +21 |
Rather than simply noting that High Point was superior, next year’s model must account for how the Panthers build leads.
High Point’s elite offensive efficiency, aggressive turnover generation, and relentless pace caused games to snowball in ways standard possession models tend to smooth out.
2. Decoupling Metrics From Narrative
Our four straight-up misses highlight another key insight. The UAB and Southern Illinois games were genuine evaluation errors.
However, the UNC Asheville and second Longwood games suggest the model occasionally overcorrected for external noise—such as venue changes, recent form, or rivalry history—at the expense of baseline team strength.
The Path Forward: Toothsayer 2.0
The biggest takeaway from Year 1 is simple: The process is sound. The Toothsayer consistently identified High Point’s superiority, it simply underestimated the absolute ceiling of that dominance.
To address this for the upcoming season, we are updating our workflow to a two-step modeling engine:
- Raw Efficiency & Pace Baseline: Generating an unvarnished statistical line based purely on possession metrics, offensive/defensive ratings, and net margin profiles.
- Contextual Overlay: Applying deliberate, capped adjustments for matchups, rest advantages, and home-court factors afterward.
With a tougher non-conference schedule and a competitive conference slate ahead, these refinements will allow us to deliver the most insightful, transparent, and accurate preview coverage available.
Thank you for following along during Year 1, we look forward to seeing you at tip-off for Year 2.