NBA Team Totals — Research Pipeline Ruling

Date: 2026-04-02 Scope: How team total data flows from collection through edge detection to execution Status: DEPLOYED (2026-04-02)

PIPELINE ARCHITECTURE

Pinnacle Page Load
    │
    ├── HTML Parse ──────────────> PinnacleMatchup[] (ML, spread, total)
    │                                   │
    │                                   ▼
    │                           storePinnacleRows()
    │                                   │
    │                                   ▼
    │                           sports_odds_snapshots
    │                           (source='pinnacle')
    │
    └── Arcadia API Intercept ──> PinnacleTeamTotal[] (per-team line + odds)
                                        │
                                        ▼
                                storePinnacleTeamTotalRows()
                                        │
                                        ▼
                                sports_odds_snapshots
                                (source='pinnacle-teamtotal')

Kalshi API Pull
    │
    ▼
KalshiAltLine[] (marketType='teamTotal', teamCode='BOS')
    │
    ▼
storeKalshiRows()
    │
    ▼
sports_odds_snapshots (source='kalshi')

Edge Scanner (scanForEdges)
    │
    ├── Pinnacle spread + total ──> impliedScores() ──> per-team expected score
    │
    ├── Normal curve (σ ≈ 9.9) ──> P(team > threshold)
    │
    ├── Compare model prob vs Kalshi execution price
    │
    ├── Apply 7% Kalshi fee
    │
    └── Store edge in sports_edges (marketType='teamTotal')

DATA COLLECTION SCHEDULE

Pinnacle Team Totals

Team totals are collected in the same browser session as regular Pinnacle odds — no extra Chromium launches.

Kalshi Team Total Markets

ARCADIA API DETAILS

How It Works

Pinnacle's React SPA fetches odds from guest.api.arcadia.pinnacle.com. During page load:

1. Puppeteer intercepts all responses from the Arcadia domain
2. Captures the x-api-key from request headers (reusable)
3. Captures the league ID from URL path
4. Captures matchup data (team names, participant IDs, alignments)
5. Captures straight market data (all market types)

Identifying Team Totals

In the straight markets response:

The participantId field is what distinguishes team totals from game totals.

Fallback

If the API intercept fails (no data captured during page load), the scraper:

1. Uses the captured x-api-key + leagueId to call the API directly via page.evaluate()
2. If that also fails, falls back to scrapePinnacleOdds() (regular scraper, no team totals)

STORAGE FORMAT

sports_odds_snapshots — Pinnacle Team Totals

The game_id format Away@Home::TT:TeamName encodes the team identity so the edge scanner can match it to the correct Kalshi alt lines.

sports_edges — Team Total Edges

TEAM CODE MATCHING

The pipeline maps between three naming systems:

Pinnacle:  "Boston Celtics"
Kalshi:    "BOS"
ESPN:      "Boston Celtics"

Mapping uses standardize() from team-codes.ts + findKalshiCode():

1. Pinnacle team name → standardized name via standardize()
2. Standardized name → Kalshi 3-letter code via findKalshiCode()
3. Kalshi alt line teamCode matched against Kalshi codes for home/away

EDGE SCANNER INTEGRATION

Input

scanForEdges() receives:

• pinnacleSnapshot: PinnacleMarket[] — spread + total per game
• kalshiSnapshot: KalshiAltLine[] — includes marketType='teamTotal' lines with teamCode

Team Total Scan Block

1. Filter Kalshi lines: marketType === 'teamTotal' && teamCode
2. Group by gameId
3. For each game, find Pinnacle total + spread
4. Derive impliedScores() → home/away expected scores
5. For each team's alt lines:
   • Match teamCode to home/away via Kalshi code lookup
   • Build Normal curve: mean = teamImplied, σ = gameSigma / sqrt(2)
   • Calculate P(score > threshold) via Normal CDF exceedance
   • Compare against Kalshi execution price
   • Apply 7% fee, tail confidence, spread filter
   • Store if executable

Dedup

Each ticker is deduped per scan type per day:

dedupKey = `${alt.ticker}::shin`

FRESHNESS MONITORING

SETTLEMENT

Team total settlement follows the same pattern as game totals:

1. Final score obtained from ESPN scoreboard API
2. Each team's final score compared to each alt line threshold
3. Over = team scored MORE than threshold → YES wins
4. Under = team scored LESS than threshold → NO wins
5. Push = team scored EXACTLY threshold → contract voids (Kalshi rules)

Settlement is sport-agnostic — the existing settleGameEdges() function handles marketType='teamTotal' alongside other types.

RISK FACTORS

EXPANSION ROADMAP

MLB team totals can be activated by adding 'mlb' to the combined scraper path in collectPinnacle().