Bet Tracker

Bet Tracker

Default data tool: PuckAPI (puckapi-tool). Use get_games (5 credits) to pull outcomes. Use get_line_movement (25 credits per game) for CLV tracking -- batch at day's end. For logging and metrics: local CSV or SQLite -- no credits consumed.

You are the honesty mechanism in the methodology chain. Your goal is to answer "do I actually have an edge right now?" with real numbers, not historical backtesting confidence. Without this skill, everything else -- model building, edge detection, daily cards -- is academic. The feedback loop that makes it matter.

When to Use

• User wants to log a bet they placed (or are about to place)
• User asks whether their model has an edge based on live performance
• User asks about closing line value, CLV, or whether the market agreed with them
• User wants to see their running ROI, win rate, drawdown, or P&L
• User asks if their track record is statistically significant
• User wants to know when to stop or pause betting

When NOT to Use

• Finding today's edges before betting -- see daily-card or edge-detection
• Historical backtesting with past data before going live -- see backtesting
• Rebuilding or retraining the model -- see model-building
• Checking calibration of model probabilities -- see probability-calibration

Commands Available

Commands That Do NOT Exist

Initial Assessment

Before starting:

1. Is the user logging a new bet, resolving existing bets, or reviewing performance metrics?
2. Where is the bet log stored? (Default: bets.csv in working directory; or SQLite bets.db)
3. Does the user track CLV? If yes, get_line_movement will be needed at day's end (25 credits/game).

Storage Schema

Recommend CSV for simplicity, SQLite for scale (200+ bets).

CSV schema (bets.csv):

date,game_id,sport,bet_type,selection,model_prob,market_odds,closing_odds,stake_pct,stake_units,result,pnl_units,clv,book,notes

Field definitions:

• date -- ISO 8601 (YYYY-MM-DD)
• game_id -- from get_games output; links to outcome resolution
• sport -- NHL (or other sport if applicable)
• bet_type -- ML, spread, total, prop
• selection -- Home ML, Away ML, Over 6.0, etc.
• model_prob -- decimal (0.0-1.0)
• market_odds -- American format at time of bet (-110, +135, etc.)
• closing_odds -- American format at game time (fill after close for CLV)
• stake_pct -- percentage of bankroll (e.g., 0.021 = 2.1%)
• stake_units -- units staked (e.g., 1.0 unit)
• result -- W, L, P (push), or OPEN
• pnl_units -- profit/loss in units (+0.91 for win at -110, -1.0 for loss)
• clv -- closing odds - market odds (positive = beat the close)
• book -- DraftKings, FanDuel, Pinnacle, etc.
• notes -- free text

How It Works

Step 1: Log a Bet

When user says they placed a bet, write a row immediately with result=OPEN.

Do not wait for the outcome. Capture at bet time: model_prob, market_odds, stake_pct, book.

Closing odds and result get filled in later.

Step 2: Resolve Open Bets

Call get_games with the relevant dates and teams. Match by game_id or team names + date.

Update result (W/L/P) and compute pnl_units:

• Win at American odds +XXX: pnl = stake_units * (odds/100)
• Win at American odds -XXX: pnl = stake_units * (100/abs(odds))
• Loss: pnl = -stake_units
• Push: pnl = 0

Step 3: Compute CLV

At end of day (not before bet placement), call get_line_movement for each game bet.

# For a bet on Team A ML:
# clv = closing_team_a_odds - bet_team_a_odds (American)
# Positive CLV: you got better odds than close (beat the market)
# Negative CLV: market moved away from your number

Convert American to decimal for averaging across books:

def american_to_decimal(odds):
    if odds > 0:
        return (odds / 100) + 1
    else:
        return (100 / abs(odds)) + 1

clv_decimal = american_to_decimal(closing_odds) - american_to_decimal(bet_odds)

CLV interpretation:

• Positive sustained CLV: model has real signal; the market confirmed your reads
• CLV near zero: you're betting at market consensus; no information advantage
• Negative CLV: market consistently disagrees with your model

Step 4: Running Metrics

Compute from the bet log at any point:

Win rate:

win_rate = wins / (wins + losses)  # exclude pushes

ROI in units:

roi = total_pnl_units / total_stake_units

Breakeven rate at average odds:

# For -110 average odds: breakeven = 110/210 = 52.38%
avg_decimal = mean([american_to_decimal(o) for o in market_odds])
breakeven_rate = 1 / avg_decimal

Average CLV:

avg_clv = mean(clv_column)  # in decimal odds units

Drawdown:

cumulative_pnl = pnl_units.cumsum()
peak = cumulative_pnl.cummax()
drawdown = cumulative_pnl - peak  # negative values = drawdown
max_drawdown = drawdown.min()
current_drawdown = drawdown.iloc[-1]

Step 5: Edge Decay Detection

Plot CLV and ROI as rolling 20-bet windows over time.

Negative CLV for 3 consecutive weeks = pause, not persevere. Do not interpret a losing streak as variance without checking CLV first.

If CLV is positive but ROI is negative: good process, bad luck. Continue. If CLV is negative and ROI is negative: the model has no real edge. Pause and audit with backtesting. If CLV is positive and ROI is positive: keep going.

Step 6: Significance Testing

Chi-squared test on win rate:

from scipy.stats import binom_test

# H0: true win rate = breakeven rate
# H1: true win rate > breakeven rate
p_value = binom_test(wins, n_bets, breakeven_rate, alternative='greater')

# p < 0.05 = statistically significant at 95% confidence

Minimum bets for significance by win rate (at -110 lines, breakeven = 52.38%):

State explicitly: "Your sample is N bets. At your win rate of X%, you need Y bets for statistical significance."

Never confirm an edge from fewer than 100 bets unless win rate is above 60%.

Step 7: Safety Alerts

Fire automatic alerts when:

1. Negative CLV 3 consecutive weeks: "CLV has been negative for 3 weeks. Pause and audit with backtesting before continuing."
2. Drawdown exceeds 20% of bankroll: "Current drawdown is XX%. Recommend pausing at 20%. Review model calibration."
3. Win rate drops below breakeven for 50+ bets: "Win rate of X% is below breakeven of Y% over 50+ bets. Auditing the model is warranted."
4. Staking above 1/4 Kelly: "Recorded stake of X% exceeds 1/4 Kelly recommendation of Y%. Confirm this is intentional."

Credit Usage

Anti-patterns

Output Format

Dashboard view:

Bet Tracker Dashboard -- Updated [date]

Total bets: N | Open: X | Resolved: Y
Win rate: XX.X% (need X more for significance at current rate)
Breakeven rate: XX.X% (at avg odds)

ROI: +X.XX units (+XX.X%)
Current drawdown: -X.X units (-X.X% of bankroll)
Max drawdown (historical): -X.X units

CLV (avg): +X.X cents [POSITIVE / NEGATIVE]
CLV (last 20 bets): +X.X cents [TRENDING UP / DOWN]

Status: [EDGE CONFIRMED / EDGE UNCERTAIN / PAUSE RECOMMENDED]
[Safety alert if triggered]

Single bet log:

Logged: [Date] | [Game] | [Bet Type] | [Selection]
Model: XX.X% | Odds: [+/-XXX] | Stake: X.X%
Status: OPEN (result pending)

Resolved bet:

Resolved: [Date] | [Game] | [Selection]
Result: [W/L/P] | P&L: [+/-X.XX units]
CLV: [+/-X cents] | [Beat the close / Didn't beat the close]
Running ROI: [+/-XX.X%] over N bets

What to Do Next