MLB Player Props Desk Construction Spec

Status: IN PROGRESS Started: April 7, 2026 Last Updated: April 7, 2026 (Phase 1+2 complete) Mirrors: MLB Desk Construction Spec

1. Data Sources

Game-Day Props (6/8/10/2/6 ET + 1 min before close)

Season Props (adaptive schedule)

Awards (adaptive schedule)

Stats (shared with game desk — read only)

Edge Detection (fires :10 past hour, after data collection)

2. Databases (7 files + 4 shared)

Game-Day Props

• fd-mlb-props.db — FanDuel prop lines (sharp anchor). Table: fd_mlb_prop_lines. Columns: player_name, player_id, market (pitcher_strikeouts, batter_hits, etc.), line, over_price, under_price, game_id, home_team, away_team, game_date, game_time, scan_type, captured_at. Same-day only, scan window gated.
• dk-mlb-props.db — DraftKings prop lines (cross-validation). Table: dk_mlb_prop_lines. Same schema as FD. Secondary table other_book_prop_lines for BetRivers/WilliamHill/BetOnline when available.
• kalshi-mlb-props.db — Kalshi player prop contracts (already exists, 572K rows). Table: kalshi_mlb_props. Columns: ticker, player_name, prop_type, threshold, yes_bid, yes_ask, volume, snapshot_type, captured_at.

Season Props

• kalshi-mlb-prop-futures.db — Season-long player prop futures. Table: kalshi_mlb_prop_futures. Columns: ticker, player_name, prop_type (season_hr, season_stat, season_stat_count), threshold, yes_bid, yes_ask, volume, snapshot_type, captured_at.

Awards

• kalshi-mlb-awards.db — Individual award futures. Table: kalshi_mlb_awards. Columns: ticker, award_type (mvp, cy_young, roy, moty, cpoty, reliever_oty, hank_aaron, ws_mvp, hr_derby), league (AL/NL), player_name, yes_bid, yes_ask, volume, snapshot_type, captured_at.

Edge Detection & Models

• mlb-prop-edges.db — All prop edge detection output. Tables:
  • player_prop_edges — Edge scanner results per player per prop. Columns: player_name, prop_type, side, line, anchor_line, fd_prob, dk_prob, kalshi_price, execution_price, raw_edge, net_edge, confidence_tier, executable, distribution_type, player_sigma, scan_type, detected_at, actual_outcome, settled_at, closing_price, clv.
  • mlb_prop_probability_curves — Precomputed probability curves per player per prop type. Columns: scan_type, game_date, game_time, player_name, player_id, prop_type, threshold, fd_anchor, dk_anchor, fd_yes, fd_no, edge_yes, edge_no, kalshi_yes, kalshi_no, rung, captured_at.
  • mlb_prop_steam — Steam detection signals. Columns: player_name, market, signal_type, bookmaker, old_line, new_line, line_move, old_price, new_price, price_move, detected_at.
  • prop_edge_summary — Per-day per-prop-type edge counts and averages.
• mlb-prop-model.db — Player baselines and matchup context. Tables:
  • prop_baselines — EWMA baselines per player per stat. Columns: player_name, player_id, stat_type, ewma_value, ewma_span, games_used, last_updated.
  • prop_matchup_context — Per-game matchup adjustments. Columns: player_name, game_date, opponent, opp_sp_name, opp_sp_hand, opp_sp_k9, park_hr_factor, park_k_factor, umpire_k_adj, platoon_factor, fatigue_factor, lineup_position, projected_pa.
  • prop_player_variance — Per-player distribution parameters. Columns: player_name, stat_type, mean, sigma, cv, dispersion_ratio, distribution_type (normal/negbin/zip), games_sampled.
  • prop_correlations — Cross-prop correlation matrix. Columns: player_name, stat_a, stat_b, correlation, games_sampled.

Shared (read-only from props desk)

• mlb-stats.db — Schedule, lineups, standings
• mlb-pitching.db — SP game logs, baselines, bullpen
• mlb-batting.db — Batter game logs, baselines, platoons
• mlb-model.db — Park factors, weather multipliers
• mlb-weather.db — Game-day weather
• pinnacle-mlb-futures.db — Sharp anchor for award futures (if Pinnacle lists them)

3. Schedules

Game-Day Props Timeline (6 AM, 8 AM, 10 AM, 2 PM, 6 PM ET)

Season Props & Awards Adaptive Schedule

Stats (9 AM ET daily — 3 staggered groups)

Props desk reads from the shared stats databases populated by the game desk's 3 scraper groups. No separate stats scrape needed.

4. Edge Detection

Game-Day Prop Edges

The prop edge scanner compares FanDuel de-vigged fair value against Kalshi prop prices:

1. FanDuel provides the sharp anchor (over/under line + prices for each prop)
2. De-vig FD's over/under to get the true probability at the posted line
3. Build a probability curve from the FD anchor using the appropriate distribution:
   • Pitcher strikeouts — Negative Binomial (overdispersed counts)
   • Batter hits — Negative Binomial
   • Batter home runs — Zero-Inflated Poisson (rare event, many 0-HR games)
   • Batter total bases — Negative Binomial
   • H+R+RBI combo — Normal approximation (sum of multiple stats)
   • Stolen bases — Zero-Inflated Negative Binomial (very rare event)
   • Runs scored — Poisson
   • Pitcher outs — Normal
4. Each Kalshi threshold is compared: if Kalshi diverges from FD fair value by more than the fee (7%), it's an edge
5. DraftKings serves as cross-validation: if FD and DK agree within 2%, confidence is HIGH
6. Edges tagged with player_sigma, matchup context, scan_type, confidence tier

Prop Probability Curves

Precomputed per player per prop type per game. Stored in mlb_prop_probability_curves table (mirrors mlb_probability_curves from game desk):

• FD anchor calibrates the distribution (like Pinnacle calibrates game curves)
• Edge curve uses player-specific baselines + matchup adjustments
• Kalshi price is what we compare against
• One row per player per threshold per scan

Matchup Context Adjustments

Applied to each prop curve before edge detection:

• Opposing SP quality — K rate, HR rate, WHIP vs league average
• Platoon splits — LHB vs LHP performance differential
• Park factors — HR factor (L/R split), K factor per stadium
• Umpire K-zone — Strike zone size affects pitcher K and batter K
• Fatigue — Back-to-back games, travel, day games after night
• Lineup position — Projected PA count affects counting stats
• Catcher framing — Affects pitcher K rate by ~0.5 K/game

Prop Steam Detection

Detects sharp line movement across books:

• FD line move > 0.5 units within 1 hour
• FD + DK move in same direction (consensus signal)
• Prop removed entirely (player scratched or lineup change)
• Prop added (player confirmed in lineup)

Season Props Edge Detection (future)

Compare Kalshi season HR/stat totals against projection systems. Anchor from PECOTA, ZiPS, or Steamer projections. Gap between Kalshi price and projection-implied probability = edge.

Awards Edge Detection (future)

Compare Kalshi award futures against Pinnacle sharp odds (when available) or consensus power rankings. WAR projections as secondary anchor. Track vote share momentum through season.

5. Dashboard

MLB Player Props Desk Groups

• Edge Detection — Prop edge scanner by type (K, hits, HR, TB, H+R+RBI), steam detection, edge summary
• Game-Day Props — FD prop lines, DK prop lines, Kalshi props (per prop type sub-views)
• Season Props — Season HR, stat milestones, stat counts
• Awards — MVP (AL/NL), Cy Young, ROY, MOTY, CPOTY, Reliever OTY, Hank Aaron, WS MVP, HR Derby
• Player Data — Batter game logs, batter baselines, SP game logs, SP baselines, platoon splits, player crosswalk
• Game Context — Matchup context, park factors, weather, umpire scorecards, lineups, schedule
• Computed Analytics — Player variance/CV, prop correlations, prop probability curves, EWMA baselines, matchup adjustments

Sub-Views per Prop Type

Each prop type should have its own filtered edge view:

• Pitcher Strikeouts (book anchor / model anchor)
• Batter Hits (book anchor / model anchor)
• Batter Home Runs (book anchor / model anchor)
• Batter Total Bases (book anchor / model anchor)
• H+R+RBI Combo (book anchor / model anchor)

6. Freshness Rules

Freshness Tracker

Same passive poller as game desk (every 5 min). Checks MAX(timestamp) per table.

Thresholds

Alerts

Same as game desk: Telegram alert after 2 consecutive misses. Consolidated message.

7. Data Quality Rules

1. No live/in-game data — Only pre-game props. Games that have started are skipped.
2. No settled prices — Kalshi bid <=5 or bid >=95 filtered out.
3. Same-day only — Game-day prop databases only store today's props.
4. Scan window tagging — Every row tagged with scan_type (6am/8am/10am/2pm/6pm/close).
5. Player name normalization — Standardize names across FD, DK, Kalshi, MLB Stats API (use player crosswalk).
6. Minimum volume filter — Kalshi props with 0 volume and spread >30 cents are low-confidence.
7. FD/DK agreement check — Edges where FD and DK disagree by >10% are flagged LOW confidence.
8. Rate limiting — Odds API has monthly quota (paid tier). Space calls 1s apart.
9. Closing snapshots — Fire 1 minute before game start for each game (staggered, not all at once).
10. Expiry dates — Season prop/award databases have expiry handling. Scanning stops after season ends.

8. Isolation Checklist Status

9. Build Priority

Phase 1: Foundation — COMPLETE (Apr 7, 2026)

1. Split FD/DK from shared mlb-prop-lines.db into fd-mlb-props.db and dk-mlb-props.db DONE
2. Fix player ID crosswalk — parse Kalshi tickers, extract player names DONE
3. Kill legacy mlb-props-edge-scanner, consolidate to unified scanner DONE
4. Fix matchup context refresh (broken since Apr 3) — added to Group 3 cron DONE
5. Split season props from kalshi-mlb-props.db into kalshi-mlb-prop-futures.db DONE
6. Split awards from kalshi-mlb-futures.db into kalshi-mlb-awards.db DONE

Phase 2: Data Quality — MOSTLY COMPLETE (Apr 7, 2026)

7. Isolate prop edges from research-pipeline.db into mlb-prop-edges.db DONE
8. Backfill 3 seasons of game logs (2023-2025) for stable EWMA baselines — PENDING (heavy one-time job)
9. Wire batter baselines into dashboard views DONE (already tracked)
10. Refresh umpire K-zone pipeline into matchup context DONE (already in matchup builder)

Phase 3: Models

11. Build prop probability curves (mirrors mlb_probability_curves pattern)
12. Player-specific distribution parameters (per-player sigma/CV)
13. Cross-prop correlation matrix
14. Manager hook probability model (pitcher K/outs)
15. Catcher framing integration

Phase 4: Polish

16. Prop-specific freshness thresholds
17. Dashboard views for all 7 groups
18. Isolation checklist — mark all items done
19. Awards edge detection model
20. Season props edge detection model

Document Control

This spec mirrors the MLB Desk Construction Spec structure. Updated as the desk is built. Final version when all checklist items pass.