The Model
What the market
hasn't priced in.
The Fine Print model doesn't predict who wins. It finds games where the market line is wrong — where structural factors like pace mismatch, travel burden, referee tendencies, and team form create a measurable gap between the posted spread and what the data says should happen.
We only publish picks when that gap is large enough to matter. Everything else is noise.
The Philosophy
What We Believe
The market is efficient — but not perfectly.
Sportsbooks set lines that reflect public betting action, not pure probability. Public money flows toward recognizable teams, recent winners, and narrative-driven favorites. That creates systematic gaps — spots where the line moves away from the true edge and disciplined data wins.
What We Don't Do
We don't chase trends or publish every game.
Most pick services publish 6-10 picks nightly regardless of signal quality. The Fine Print model publishes only when strong divergence exists between our composite score and the market line. Most nights that's 1-2 games. Some nights it's zero. That discipline is what produces a 60%+ record over a full season.
How A Pick Is Generated
01
Data Collection
Each morning the pipeline scrapes game results, injury reports, referee assignments, and current betting lines. Updated daily at 10am ET.
02
Signal Scoring
Each game is scored across seven validated signals — pace, form, quality, timezone, streaks, referee tendency, and rest. Each signal is standardized against league distributions.
03
Composite Score
Signals are combined into a single composite score. A sigmoid function converts this to an implied win probability for comparison against the market.
04
Market Comparison
The model's implied probability is compared against the market's implied probability from the spread. Only when the gap exceeds a strong threshold is a pick published.
The Signals
⚡
Pace Mismatch
Game Environment
✓ Strong Signal
Every NBA team has a natural pace — possessions per game — that reflects their style of play. When a fast-paced away team enters a slow home team's building, the home team controls the tempo. Slow home teams dictate pace in their own arena, forcing the faster away team into an uncomfortable rhythm they rarely escape.
Validated Finding
Slow home teams hosting fast away teams win at 65–73% historically across 4 seasons. This signal was inverted in early model versions — catching and correcting it was one of the most impactful fixes of the 2025-26 season audit.
65–73%
Home win rate
slow home vs fast away
4 seasons · consistent
📈
Form Gap
Recent Performance
✓ Strong Signal
Form measures how a team is performing relative to their own season average over recent games. The model compares both teams' recent form and calculates the gap in standard deviations. A large form gap is one of the strongest predictors of spread outcome in the entire model.
Validated Finding
When the form gap reaches -3 standard deviations in the home team's favor, home win rate hits 88.9%. The gradient is monotone — the larger the form gap, the more predictable the outcome. This is the highest-weighted signal in the composite.
88.9%
Home win rate
at -3 SD form gap
monotone gradient
🗺️
Away Quality
Team Strength
✓ Strong Signal
Not all road teams are equal. The model measures away team quality using road win percentage — not overall record — compared against the home team's home win percentage. This catches situations where a team looks competitive on paper but is genuinely poor on the road.
Validated Finding
Using road win% vs home win% instead of overall records significantly improved accuracy on games with large home/away splits. A team that is 9-27 on the road is a fundamentally different opponent than their 25-49 overall record suggests.
Road
Win% used
not overall record
context-aware split
👔
Referee Tendency
Officials
~ Moderate Signal
Specific referees have measurable home bias — they call fouls at different rates depending on whether the home or away team is involved. This is about unconscious crowd influence documented in officiating research for decades. The model tracks each referee's home win rate across 3+ seasons and applies it when their assignment is confirmed.
Validated Finding
Most referees are close to neutral. However, outlier referees with persistent multi-season home bias represent measurable edge. The weight was reduced in the 2026 audit after confirming the signal is real but weaker than initially modeled.
Real
Signal confirmed
outlier referees
3+ season sample
💤
Rest Differential
Fatigue
→ Tiebreaker Only
Rest differential measures the gap in days off between the two teams. The market already prices rest differential aggressively, which means bettors acting on rest alone are paying for information the line already reflects.
Validated Finding
Across all 4 seasons of data, rest differential is confirmed noise as a standalone betting signal. The model retains it at a very low weight as a tiebreaker only — it is never the reason a pick is published.
Noise
Standalone signal
market-priced
confirmed 4 seasons
Market Divergence — The Final Filter
Skip
Low
Model and market are aligned. No meaningful edge exists. No pick is published regardless of which team the model favors. Most games on any given night fall into this category.
Moderate
Watch
Some divergence exists but below the strong threshold. Listed in the model report as a game worth monitoring. The signal is present but not strong enough to publish a pick with confidence.
Strong
59.3%
Strong divergence on spreads of 10 or fewer points. This is the only tier where a pick is published. Current season record: 51-35 on strong divergence picks. Updated nightly on the NBA record page.
Model Transparency
We ran a full audit mid-season and published what we found.
In March 2026 we ran an 11-query data scientist audit of every signal in the model. We found a sign error in the pace mismatch coefficient that had been generating wrong-direction picks all season — the signal was real but inverted. We corrected it, documented it, and updated the model. We don't hide mistakes. We fix them and move forward. That's what a data-driven approach actually looks like.
Common Questions
How many picks does the model publish per night?
On average 1-3 picks on a full slate night. Some nights zero if no game meets the strong divergence threshold. The model never forces picks to fill a quota — that discipline is what separates 60%+ from 52%.
Does the model work in the playoffs?
The playoff model is still developing. When the same two teams play 7 games in a row, the market adjusts much faster than in the regular season. We publish playoff picks with a clearly labeled lower confidence designation so subscribers can calibrate accordingly.
Why is the over/under model paused?
The O/U model showed a 57.1% hit rate on 49 games at the primary threshold — not enough edge to publish with confidence yet. The underlying signals are promising but we need a larger sample. We'd rather wait than publish a model that isn't ready.
Can I access the raw signal data?
Yes. The Fine Print Analytics API provides programmatic access to pace profiles, referee tendencies, travel data, timezone lag, player streaks, and ATS results. Available on RapidAPI with plans starting free.
Is the pick record verified?
Yes. The public pick record page updates automatically each morning after game results are finalized. Every pick is logged with the date, matchup, line, and result — generated directly from the same database that powers the model. No manual entry, no cherry-picking.
See the model in action.
View the full verified pick record or subscribe to receive picks before tip-off.