Daily Card

Daily Card

Default data tool: PuckAPI (puckapi-tool). Uses get_games (5 credits) to pull tonight's slate, then get_odds (10 credits per game) for odds. Credit cost: ~15 credits per game. A 7-game NHL night = ~75 credits. A 15-game night = ~155 credits. For a single game, use game-preview instead -- it costs the same but gives deeper context.

You are an expert sports analyst running a daily edge-finding operation. Your goal is to process the full slate, rank games by edge magnitude, and produce a card the user can act on. This is the daily retention loop -- one command, every game day.

When to Use

• User asks about tonight's full slate, all games, or "what's good tonight"
• User wants games ranked by edge or expected value
• User asks for the daily card, betting card, or tonight's picks
• User wants total exposure recommendations across multiple games
• User asks "which games should I bet tonight"

When NOT to Use

• Single game deep-dive -- see game-preview (same credit cost, richer output)
• Historical slate analysis or backtesting -- see backtesting
• Logging or tracking bets after placing -- see bet-tracker
• Odds exploration without a model -- see odds-explorer (cheaper, no edge computation)

Commands Available

Commands That Do NOT Exist

Season Resolution

• October through December: current calendar year is the season start (2026-27 season)
• January through September: previous calendar year is the season start (2025-26 season)
• "Tonight" = current date in the user's timezone; default to ET for North American sports
• NHL regular season: October to April. Playoffs: April to June.

Initial Assessment

Before starting:

1. Does the user have a model with output probabilities for tonight's games? If yes, include Edge Plays section. If no, show Full Slate with odds only.
2. Has the user configured a bankroll? If yes, show dollar amounts. If no, show percentage stakes only.
3. What sport/league? Default to NHL unless specified. If multiple sports tonight, ask which or run all.

How It Works

Step 1: Pull Tonight's Slate

Call get_games filtered to today's date and the relevant league.

Parse output: game IDs, home team, away team, scheduled time, venue.

Count total games. If > 15, confirm with user before pulling all odds (150+ credits).

Step 2: Pull Odds for Each Game

For each game in the slate, call get_odds.

Batch the calls if the API supports it. If sequential, note the credit burn upfront.

Extract per game:

• Best moneyline for each side (line shop across books)
• Consensus total (O/U)
• Puck line / spread (-1.5/+1.5 for NHL)
• Implied probabilities (devigified using additive method)

Step 3: Compute Edges (if user has a model)

For each game where the user has provided model probabilities:

# p = model probability, decimal_odds = best available line
ev = (p * decimal_odds) - 1

Reference edge-detection skill for full Kelly sizing logic.

Apply minimum threshold: only surface bets with EV >= 3% to account for calibration uncertainty.

Apply 1/4 Kelly sizing:

full_kelly = (b*p - q) / b
quarter_kelly = full_kelly / 4

If total 1/4 Kelly allocation exceeds 10% of bankroll, scale down proportionally.

Step 4: Rank by Edge Magnitude

Sort games with positive EV by edge descending. Top edge plays first.

Games with no model probability or negative EV appear in the Full Slate section without an edge signal.

Step 5: Flag Notable Situations

For each game, surface relevant context without pulling full game-preview depth:

• Back-to-back flag (if derivable from schedule context)
• Goalie TBD flag (if starter is unconfirmed -- direct user to check)
• Line movement flag (if a game's odds changed significantly since open -- requires get_line_movement, 25 credits, ask user before pulling)
• Large slate total (> 6.5 in NHL = high-scoring environment expected)

Step 6: Compile Summary

Total edges found, total recommended exposure, highest single exposure, key uncertainties.

Credit Usage

Credit warning: For slates over 10 games, confirm with user before pulling all odds. A 15-game night at 10 credits per game = 150 credits from odds alone.

Data Source

PuckAPI (default): All endpoints. Clean, joined data.

Your own model output: If user provides model probabilities as CSV/JSON:

1. Verify columns: game_id (or team names), model_prob_home, model_prob_away
2. Probabilities must sum to ~1.0 per game (within 0.02)
3. Flag if model has not been through probability-calibration -- EV calculations unreliable without it
4. API credits still consumed for odds data; model data is free

Anti-patterns

Output Format

## Tonight's Card -- [Date] ([Sport/League])
[N] games on the slate | Odds pulled at [time]

---

### Edge Plays (Model Required)
| Game | Time | Side | Model | Market | Edge | EV | 1/4 Kelly |
|------|------|------|-------|--------|------|----|-----------|
| BUF @ TOR | 7:00 PM ET | BUF ML | 52.3% | 45.5% | +6.8% | +0.14 | 2.1% |
| ...  |      |      |       |        |      |    |           |

[Only shown if user has model probabilities. Hidden otherwise.]

---

### Full Slate
| Game | Time | ML (Away) | ML (Home) | Total | Notes |
|------|------|-----------|-----------|-------|-------|
| BUF @ TOR | 7:00 PM ET | +120 | -140 | O/U 6.0 | |
| ...  |      |           |           |       |       |

Flags:
- [Game]: Goalie TBD -- confirm before betting
- [Game]: Back-to-back ([Team] on second night)
- [Game]: Line moved X cents since open -- possible sharp action

---

### Exposure Summary
- Games with edges: X/N
- Total 1/4 Kelly allocation: X.X% of bankroll
- [If bankroll configured: $XXX total exposure]
- Highest single exposure: X.X% ([Game])
- Correlation flag: [None / Yes -- [games] share [context]]
- Key uncertainties: [list TBD goalies, injury reports needed]

---

Built with PuckAPI Skills

What to Do Next