NBA Team Totals — Data Audit Council Ruling

Date: 2026-04-02 Process: Architectural review — data sources, probability model, edge detection methodology Scope: NBA team totals market (individual team scoring over/under) Status: BUILT & DEPLOYED (2026-04-02)

WHAT IS A TEAM TOTAL?

A team total is a bet on how many points a specific team will score in a game (e.g., Lakers Over/Under 112.5). This is distinct from the game total which is the combined score of both teams.

Why it matters: Team totals are less efficiently priced than game totals because they require team-specific modeling rather than aggregate game modeling. This creates edge opportunities.

Kalshi series: KXNBATEAMTOTAL

• Event format: KXNBATEAMTOTAL-26APR03BOSMIL (date + away + home)
• Market format: KXNBATEAMTOTAL-26APR03BOSMIL-BOS116 (team + line threshold)
• ~9 alt lines per team at 3-point intervals

DATA SOURCES

Primary Anchor: Pinnacle Team Totals (NEW)

Secondary Anchor: Derived from Pinnacle Game Lines

When Pinnacle team total odds are unavailable, team totals are derived:

Home team total = (game_total + spread) / 2
Away team total = (game_total - spread) / 2

Example: Game total 225.5, home -5.5
  Home TT = (225.5 + 5.5) / 2 = 115.5
  Away TT = (225.5 - 5.5) / 2 = 110.0

This derivation uses Pinnacle's spread and game total — both already scraped on adaptive cadence.

Existing Data Sources (Already Collected)

PROBABILITY MODEL

Distribution: Normal

NBA team scoring is approximately normally distributed. The edge scanner uses:

teamSigma = gameSigma / sqrt(2)

Where:
  gameSigma = 14 (NBA game total standard deviation)
  teamSigma = 14 / sqrt(2) ≈ 9.9

Model probability = Normal CDF exceedance at threshold
  P(team scores > threshold) = 1 - Phi((threshold - teamImplied) / teamSigma)

Anchor Selection

The team implied score comes from:

1. Pinnacle team total line (preferred — direct market price)
2. Derived from spread + game total (fallback — mathematical identity)

Both are valid. The direct Pinnacle line is preferred because it includes market information about team-specific factors (injuries, pace matchups, rest) that the derivation doesn't capture.

De-Vig Method

Team total odds from Pinnacle are de-vigged using the Shin method (same as game totals):

Shin de-vig removes the bookmaker margin while preserving the
favorite-longshot bias inherent in real odds.

The de-vigged over probability becomes the model's fair value anchor.

EDGE DETECTION METHODOLOGY

Scan Flow

1. Pinnacle team total odds → de-vig → fair over/under probability
2. Build Normal curve around team implied score (σ ≈ 9.9)
3. For each Kalshi alt line:
   a. Calculate model probability at that threshold
   b. Compare to Kalshi mid-price
   c. Calculate raw edge = model_prob - execution_price
   d. Subtract Kalshi fee (7% of profit)
   e. If net_edge > min_threshold → flag as executable
4. Store in sports_edges table with marketType='teamTotal'

Edge Filters

Rung Distance

Distance from anchor affects model confidence:

rungDistance = |threshold - teamImplied| / 1.5

Thresholds far from the implied score have lower model confidence
because the normal approximation degrades in the tails.

CUSTOM CALCULATED METRICS

1. Team Pace Delta: (Team_pace - Opponent_pace) / 2 — faster pace = higher team total
2. Defensive Rating Matchup: Team_OffRtg * (Opp_DefRtg / League_Avg_DefRtg) — offense adjusted for opponent defense
3. Rest Advantage Impact: B2B = -2.5 pts, 2+ days rest vs B2B opponent = +1.5 pts
4. Home/Away Scoring Split: NBA home teams score ~2.5 pts more on average
5. EWMA Scoring Form: Decay-weighted last 10 games scoring average (alpha 0.12)
6. Blowout Risk Flag: Teams with high scoring variance (σ > 12) have wider team total ranges
7. Pace Matchup Extreme: Both teams top-10 pace = team totals likely OVER; both bottom-10 = UNDER

NOISE TO TRIM

• IGNORE: Season-long PPG averages (use EWMA form instead)
• IGNORE: Head-to-head season series scoring (too small sample)
• COMPRESS: Quarter-by-quarter scoring to single pace metric
• STRIP: Overtime scoring from team total calculations
• WALL: All odds/prices invisible to analysts — only show matchup context

NBA-SPECIFIC FACTORS FOR TEAM TOTALS

1. Pace is king — two fast teams (Memphis vs Indiana) produce higher team totals than two slow teams (Knicks vs Cavaliers)
2. Back-to-back fatigue — reduces scoring by ~2-3 points per team
3. Garbage time inflation — blowouts inflate losing team's total in 4th quarter
4. Star player rest — load management reduces team scoring by 3-5 points when a star sits
5. 3-point variance — teams that live by the 3 have higher scoring variance
6. Altitude — Denver home games have slightly elevated totals
7. Defensive matchups — elite rim protectors (e.g., Wembanyama) suppress opponent interior scoring
8. Late-season tanking — eliminated teams may rest starters, depressing totals

INTEGRATION WITH EXISTING PIPELINE

Collector (collector.ts)

NBA Pinnacle scrapes now use scrapePinnacleWithTeamTotals() which:

• Scrapes regular matchups (ML, spread, total) from HTML
• Intercepts Arcadia API responses for team totals
• Returns both in a single browser session
• Stores team totals with source='pinnacle-teamtotal'

Edge Scanner (edge-scanner.ts)

The team total scan block (lines ~655-780):

• Filters Kalshi snapshot for marketType='teamTotal'
• Groups by gameId, matches team codes via standardize() + findKalshiCode()
• Builds Normal curve per team (σ = gameSigma / sqrt(2))
• Calculates edges, applies fee deduction, dedup, tail confidence
• Stores with marketType: 'teamTotal' and teamCode

Dashboard (desk-config.ts)

• "Edge Scanner — Team Totals" in Edge Detection group
• "Pinnacle NBA Team Totals" in Odds & Prices group
• Both wired to freshness monitoring

FUTURE IMPROVEMENTS

1. Use actual Pinnacle team total odds for de-vig instead of deriving from spread+total (deployed — Arcadia API intercept)
2. Team-specific sigma — compute per-team scoring std dev from recent games instead of league-wide 9.9
3. Pace-adjusted sigma — fast-pace matchups have higher variance
4. Correlate with player prop lines — if a star's points prop moves, team total should move proportionally
5. Live team total tracking — in-game team total edges using Kalshi live odds
6. Cross-sport expansion — NHL team totals (KXNHLTEAMTOTAL), MLB team totals (already in config), NCAAB team totals

BUILD STATUS