Tennis Desk Strategy Spec

Version: 3.0 | Date: 2026-04-03 | Status: DEPLOYED

Overview

Trade mispriced Kalshi tennis contracts using Pinnacle de-vigged fair value. Coverage: ~49 Kalshi series spanning ATP, WTA, Challenger, Grand Slams. Year-round daily action.

Edge Detection

Market	De-vig Method	Notes
Match winner (2-way)	Shin	Primary market — DEPLOYED
Game spread / Total games	Markov chain + Kalshi alt-lines	DEPLOYED — model-priced via Markov chain, compared to Kalshi prices
Set winner / Exact set score	Markov chain + Kalshi alt-lines	DEPLOYED — exact set score matrix from Markov chain
Grand Slam outright (N-way)	Monte Carlo (100K bracket sims)	DEPLOYED — SA-Elo + fatigue through rounds

Kalshi fee: 7% on profit. Gross edge ~4.3c needed to net 4c. Min edge: 4 cents after fee.

Tennis-Specific Models

Surface-Adjusted Elo (SA-Elo): 0.5 × Overall + 0.3 × Surface + 0.2 × RecentForm. Dynamic K-factor. Separate per surface (hard/clay/grass/indoor).
Serve/Return (Log5): P(A_serve) = (SPW_A × RPW_B_allowed) / TourAvgSPW
Hold Probability: p^4 + 4p^4(1-p) + 10p^4(1-p)^2 + 20p^5(1-p)^3/(1-2p(1-p))
Fatigue Score: minutes_7d + 0.5 × minutes_8to14d + travel_zones × 100
Retirement Probability: Logistic regression (age, matches_14d, MTOs, heat_index, surface)
BO5 Conversion: P(Bo5) = p^3 + 3p^3(1-p) + 6p^3(1-p)^2
H2H Adjustment: Bayesian, time-decayed (4yr ago = 0.1× last month), min N=3
Clutch Index: (Actual BP Won%) - (Expected BP Won% from RPW), mean-reverted

Edge Scanner (DEPLOYED)

Scans Pinnacle ML de-vigged vs Kalshi match prices. Uses last-name fuzzy matching for player identification. Now produces edges with correct market type labels (ml, spread, total, setscore, setwinner).

ATP scan result (latest): 132 Kalshi matched to 453 Pinnacle games, 49 edges found, 42 executable
WTA: Pending WTA-specific Pinnacle data (0 entries currently — dynamic tournament key resolver deployed)

Odds API Integration: Dynamic tournament key resolution deployed — automatically discovers active ATP/WTA tournaments from the Odds API sports endpoint. No more stale hardcoded keys.

Position Sizing

Market	Kelly Fraction	Max
Match winner	1/8	—
Game/set spreads	1/8	—
Total games	1/8	2% bankroll on extremes
Set betting	1/10	—
Exact scores	1/12	0.5% bankroll
Outrights	1/8	—
Per-match cap	—	3% bankroll

Collection Schedule

Time (ET)	Data
Every 30min	Kalshi tennis prices (49 series, ~498K snapshots)
Adaptive	Pinnacle tennis odds (dynamic tournament discovery)
Weekly	Sackmann CSV ingest, SA-Elo, surface baselines, H2H, rankings, tournaments, draws, WTA calibration, TennisAbstract
Daily	Fatigue scores, weather (24 venues)
8AM/10AM/2PM/6PM	Edge scanner (all market types)
Static	Integrity kill-switch config

Kalshi Series (49 active)

ATP Match: KXATPMATCH, KXATPCHALLENGERMATCH, KXATPDOUBLES WTA Match: KXWTAMATCH, KXWTACHALLENGERMATCH Alt Lines: KXATPGAMETOTAL, KXATPGSPREAD, KXATPGAMESPREAD Sets: KXATPSETWINNER, KXATPTOTALSETS, KXATPEXACTMATCH, KXATPEXACTSETS, KXATPANYSET Outrights: KXATPGRANDSLAM, KXWTAGRANDSLAM, KXATPGRANDSLAMFIELD Rankings: KXATP1RANK, KXATPRANK Table Tennis: KXTABLETENNIS, KXTTELITEGAME

Key Constants

Parameter	Value
SA-Elo blend	0.5 overall + 0.3 surface + 0.2 recent
Min edge	4 cents
Kalshi fee	7% on profit
Kelly fraction (match)	1/8
Min surface matches	10+ for both players
H2H min meetings	3
Elo stabilization ATP	30 initial, 50+ stable
Elo stabilization WTA	~40 matches
WTA upset rate	~38% vs ATP ~32%
Challenger Kelly	50% of normal
Integrity kill-switch	40c Pinnacle move on Challenger/ITF

Deferred (needs more data)

Item	Reason
WTA edge scanner	Needs WTA Pinnacle data flowing; dynamic resolver deployed, waiting for data

Discarded

Item	Reason
Kalshi WebSocket	REST polling adequate
Betfair exchange	Not worth integration
Press conference sentiment	Low-signal, council rejected
In-play Bayesian updating	No spec exists
OnCourt historical DB	$5K/yr, not building
Pinnacle direct feed	$1.5K/yr, not building

Source: ~/edgeclaw/results/spec-panel/sports-desk/tennis-strategy.md