NCAAB Desk Strategy Spec

Extracted from Sports Desk Merged Spec (2026-03-15)

Full merged reference: spec-final.md

OVERVIEW

The NCAAB desk detects mispricings between Kalshi alt line prices and Pinnacle sharp book fair value for college basketball games. Pinnacle is the anchor. When Kalshi's price diverges from Pinnacle's fair value, that gap is the edge.

NCAAB has higher variance than NBA — less talent parity, bigger upsets, wider blowouts. This means both more opportunity (larger mispricings) and more risk (model accuracy suffers on mismatches).

CORE EDGE DETECTION

How It Works

1. Get Pinnacle's odds for an NCAAB game (moneyline, spread, total)
2. Remove the vig to get the "fair" probability
3. Compare that fair probability to Kalshi's price for the same market
4. If Kalshi's price is off by more than 4 cents after fees, that's a tradeable edge

De-Vig Method

2-Way Markets (Spreads, Totals, Moneylines) — Multiplicative method:

implied_prob = abs(odds) / (abs(odds) + 100)  # for negative odds
implied_prob = 100 / (odds + 100)              # for positive odds

total_implied = implied_home + implied_away
novig_home = implied_home / total_implied
novig_away = implied_away / total_implied

NCAAB moneylines are 2-way (no draw). Multiplicative method works.

Kalshi Fee Impact

Kalshi charges 7% fee on profit.

win_amount = 100 - kalshi_price
fee_on_win = 0.07 * win_amount
expected_profit = (model_prob * (win_amount - fee_on_win)) - ((1 - model_prob) * kalshi_price)

Minimum net edge: 4 cents after fees.

Sweet spot: implied probabilities between 35% and 65%.

Position Sizing (Kelly Criterion)

Use 1/4 Kelly for all NCAAB bets:

bet_size = min(bankroll * full_kelly * 0.25, bankroll * 0.05)

If quarter Kelly < 0.5% of bankroll, skip.

Hard Caps:

• Maximum single bet: 5% of bankroll
• Maximum per game: 10% of bankroll
• Maximum daily NCAAB exposure: 25% of bankroll

NCAAB-SPECIFIC MODELS

Normal Distribution for Spreads and Totals

Like NBA, NCAAB scoring is high enough for Normal distribution. But NCAAB spreads are often much wider (20+ point favorites common) — model accuracy drops at extreme spreads.

Spread Model:

• Mean: Calibrated from Pinnacle's spread
• Wider variance than NBA due to talent gaps between teams

Market Priority

NCAAB-SPECIFIC EDGE FACTORS

Efficiency Ratings — Primary Edge Source

NCAAB's edge comes from advanced efficiency models that casual markets (Kalshi) don't fully price in. These models measure how many points a team scores/allows per 100 possessions, adjusted for opponent strength.

Data Sources:

KenPom and BartTorvik are the gold standard for NCAAB analytics. Their adjusted efficiency margins directly feed into spread and total predictions.

Conference Tournament / March Madness

Special considerations during tournament play:

• Neutral site games: Remove home court advantage from model
• Motivation asymmetry: Teams on the bubble play differently than locked-in seeds
• Matchup variance: Single-elimination = higher variance. Model uncertainty increases.
• Public betting bias: Casual bettors overvalue brand-name programs in March Madness, creating larger mispricings on mid-major underdogs

Talent Parity Gap

NCAAB has massive talent gaps:

• Top-25 vs unranked can produce 30+ point blowouts
• Model is most accurate for competitive matchups (spread < 10 points)
• For spreads > 15 points, increase uncertainty and reduce position size

Injury / Eligibility

Track suspensions and eligibility issues — NCAA has unique rules around academic eligibility, transfer portal, and discipline that don't exist in pro sports.

NCAAB DATA SOURCES

Rating Models

Average available models. If any is NULL, exclude silently.

NCAAB STEAM DETECTION

NCAAB has a higher steam threshold for spreads than NHL/NBA because college lines are more volatile (less liquidity, sharper moves are normal).

When steam is detected:

1. Immediately scan all Kalshi markets for that game
2. Flag edges as steam_triggered = TRUE
3. Apply 1.25x confidence multiplier (still capped at 5%)

WHEN NOT TO BET (Hard Rules)

NCAAB COLLECTION SCHEDULE

[11:00 PM ET prior night]
  - Fetch tomorrow's NCAAB schedule from ESPN API
  - Create game records

[9:00 AM ET — Early Snapshot]
  - Odds API: Pinnacle NCAAB lines
  - Kalshi: Pre-game alt line snapshot
  - Calculate no-vig probabilities, run edge scan

[11:00 AM ET (weekday) / 10:00 AM ET (weekend) — Model Collection]
  - BartTorvik/KenPom (11:06)
  - T-Rank (11:15)
  - DRatings (11:09)
  - Massey (11:18, best-effort)

[2:00 PM ET — Mid Snapshot]
  - Odds API: Pinnacle update
  - Kalshi: Mid-day snapshot
  - Calculate line velocity, flag steam moves

[30 min before game]
  - ESPN injury check
  - 30-minute pre-close polling

[10 min before game — Closing Snapshot]
  - Final Pinnacle closing lines
  - Kalshi closing price
  - Store CLV baseline

[Game start → WebSocket live mode]
[After game ends → Grade edges, update Brier scores]

SETTLEMENT AND LEARNING

Track every detected edge with: timestamp, game_id, market_type, kalshi_ticker, kalshi_price, model_fair_prob, raw_edge, net_edge, rung_distance, steam_flag, bet_placed, bet_size, pinnacle_line.

Post-Settlement:

• Record actual outcome (final score)
• Compute Brier score: (model_fair_prob - outcome)^2
• Track rolling Brier scores per market type

Monthly Review:

1. Brier score per market type
2. Profit/loss per market type
3. Edge accuracy
4. Average edge by rung distance
5. Performance by spread size bucket (<5, 5-10, 10-15, 15+)
6. Conference tournament performance vs regular season
7. Morning vs evening edges

RISK MANAGEMENT

No risk management = no live trading.

QUICK REFERENCE: KEY CONSTANTS

TODO: Upgrade Kalshi REST to WebSocket Feed

Status: NOT BUILT — add when this desk goes live for execution

Current state: All Kalshi data (prices, order books, trades) is fetched via REST API polling on cron schedules. This is fine for edge detection and monitoring, but NOT sufficient for live trade execution.

Why WebSocket matters:

• REST polling introduces lag — by the time you see a price, it may already be gone
• On tight edges (2-5 cents), even 1-2 seconds of lag can eat the entire edge
• WebSocket gives real-time order book deltas and trade notifications
• Essential for hitting specific fill prices and monitoring execution quality

What to build when ready:

1. Connect to Kalshi WebSocket feed (wss://trading-api.kalshi.com/trade-api/ws/v2)
2. Subscribe to order book channels for active tickers
3. Stream trade events for real-time freshness (replaces REST trades polling)
4. Feed live prices into execution engine for precise entry/exit
5. Update Fill Quality Monitoring to measure slippage vs quoted price

Note: Sports markets use RFQ (Request for Quote) so the visible order book is usually empty — but the trade feed still matters for freshness signals and the WebSocket is required for order submission. Weather markets DO have real visible order books where this upgrade is even more critical.

Added 2026-03-29 — upgrade REST to WS when desk moves to live execution