How Bosh Roid Works
Every signal, every grade, every unit tracked — here's exactly how our prop model operates from odds ingestion to final result.
The Pipeline — Step by Step
Sportsbook Lines Are Pulled
Multiple sportsbooks (DraftKings, FanDuel, BetMGM, and others) are queried via The Odds API throughout the day. All player prop lines — points, rebounds, assists, 3-pointers, blocks, steals, and combo props — are stored as snapshots in our database.
Each snapshot records the exact line value, over odds, under odds, and the timestamp it was captured. This enables Closing Line Value (CLV) tracking over time.
Independent Projections Are Generated
Our model generates player projections independently from the sportsbook lines. Projections factor in season averages, recent form, pace, matchup quality, and usage trends.
Each projection includes a confidence score (0–100) reflecting how reliable the underlying data is. Players with fewer than 5 games played or confidence below 50 are filtered out automatically.
Projection vs. Line Creates an Edge
The edge is calculated as: edge = projection − line_value
A positive edge means the model projects the player to beat the over line. A negative edge means the model projects an under. Each market type has a minimum edge threshold before a signal is created (e.g. points require +1.5 edge; threes only need +0.5).
The final confidence score shown on each signal blends the model confidence with the edge magnitude — larger edges that come with high-confidence projections receive the highest scores.
Odds Are Factored In
Once the direction (OVER or UNDER) is determined, we factor in the actual American odds for that side. A +120 signal is worth significantly more than a -130 signal for the same projection edge.
Expected Value is not calculated in isolation — odds pricing directly affects the ROI and units calculations used in grading and performance reporting.
Injury Status & Outlier Safeguards Applied
Real-time NBA injury reports are fetched and checked against each candidate prop edge:
- Out / Doubtful — signal is rejected entirely
- Questionable — confidence score is penalized 15-20 points
- Probable / Available — no penalty
Additionally, players with high minutes volatility in recent games (indicating unreliable role) or a history of DNPs (Did Not Play) are flagged and have confidence reduced to prevent surprises.
Alternate Lines Detected & Grouped (Ladders)
When the same player/market combination exists at multiple line values for the same side, a ladder group is created:
- Primary Play (rank 1) — best expected value or most straightforward recommendation
- Ladder Legs (rank 2+) — alternate lines grouped together for users who want to multi-leg bets
Example: Jalen Brunson Assists at Over 8.5 (-110), Over 9.5 (+210), Over 10.5 (+340) are grouped as one ladder. Only the primary is shown as a standalone top play; legs are nested underneath.
Best Sportsbook Value Is Selected
When lines from multiple sportsbooks are available for the same player/market combination, we automatically select the book offering the most favorable American oddsfor the recommended side.
Example: if DraftKings has a player's over at -120 and FanDuel has it at -105, the signal locks in FanDuel's -105 since it provides a better payout per unit risked.
Pregame Line Is Locked
At the moment a signal is created, the exact pregame line (signal_line_value) and odds (signal_odds_price) are frozen permanently.
Live odds, line moves during the game, or post-game data are never used for grading. This ensures every result is graded on exactly the line that was available when the signal was issued.
Bets Are Graded After the Final Box Score
After games conclude, our grading engine pulls finalized ESPN box score data to determine each player's actual stat. The result is graded as:
- WIN — actual stat cleared the recommended side vs the locked line
- LOSS — actual stat did not clear
- PUSH — actual stat exactly equaled the line
- VOID — player did not play (DNP) or no box score found
Grading runs automatically each night and is fully idempotent — no prop is ever graded twice.
ROI and Performance Are Tracked
Units and ROI are calculated using the actual locked American odds, not a flat -110 assumption:
- WIN (negative odds): units_won = 100 / abs(odds) × units_risked
- WIN (positive odds): units_won = odds / 100 × units_risked
- LOSS: units_won = −1 × units_risked
- PUSH / VOID: units_won = 0
ROI = (units_won / units_risked) × 100. All performance data is broken down by market type, confidence bucket, and odds bucket on the Model Performance page.
What Changed Recently
- •Injury-Aware Safeguards — Real-time NBA injury reports now applied to all prop edges; Out/Doubtful players excluded, Questionable players penalized.
- •DNP & Volatility Filtering — Recent minutes outliers and DNP history now tracked to prevent unexpected player absences.
- •Ladder Play Grouping — Multiple alternate lines (e.g., Over 8.5, 9.5, 10.5) are now grouped together with one primary play and ladder legs for multi-leg strategies.
- •Multi-Book Awareness — All available sportsbooks are tracked per signal; best odds are selected at signal creation time.
- •Steam Alerts — Line movement detection now tracks meaningful moves (≥1.0 point line changes or significant odds shifts) across all sportsbooks, alerting users to market activity.
What We Don't Do
No Live Odds Used
No Retroactive Changes
No Cherry-Picking
See it in action
Check today's signals or review the full graded results history.
Bosh Roid Fantasy Network · Prop model graded against finalized ESPN box scores only. Lines sourced from The Odds API. Steam alerts powered by sportsbook snapshot comparison.
Last updated: March 9, 2026 · Model Version: 2.6 (Injury-Aware, Ladder-Aware, Steam-Alert-Ready)