prediction-market-making

name	Prediction Market Making
description	Liquidity provision on Polymarket — spread optimization, quadratic reward maximization, adverse selection defense via real-time WebSocket monitoring, two-sided quoting with inventory management
version	1.0
format	completion-engine
workflows	3
source	MES 3.0 Deep Extraction — Polymarket docs + poly-maker + polymarket-agents + polymarket-arbitrage + ecosystem analysis (5 sources, 5,200 lines)

Prediction Market Making

Polymarket distributes $5M+/month in liquidity rewards via a quadratic scoring formula: S(v,s) = ((v-s)/v)^2 * b. This skill turns that reward pool into systematic income through two-sided quoting, adverse selection defense, and reward score optimization. The quadratic structure means a 1-cent quoter earns 3.24x the reward of a 5-cent quoter — small spread improvements yield exponential gains. Two-sided quoting earns 3x vs single-sided (c=3.0 divisor penalty). The poly-maker author confirms the bot is "unprofitable" — adverse selection destroys spread capture. Only the rewards program makes market making viable. Every workflow optimizes around that reality.

Core Principle: Reward harvesting is the business model, not spread capture. The ImMike config sets mm_enabled: false with the comment "markets too efficient." Without the rewards layer, you are paying infrastructure costs to lose money to informed traders.

---

Available Workflows

#	Workflow	Produces	Use When
market-select	Market Selection & Spread Design	Market ranking by reward pool/competition + optimal spread parameters + deployment config	Choosing which markets to quote, how much capital to allocate, what spreads to post
adverse-defense	Adverse Selection Defense	Complete defense system: news monitoring, auto-cancel/widen rules, inventory thresholds, heartbeat management, kill switch config	Building the protection layer that makes profitable market making possible
reward-optimize	Reward Optimization	Full Q chain calculation, spread/size tuning, reward vs adverse selection trade-off modeling, per-market tuning recommendations	Maximizing share of the $5M+/month reward pool after basic infrastructure is running

---

Quick Reference

Reward Formula

• Score: S(v,s) = ((v-s)/v)^2 * b — quadratic, tighter is exponentially better
• Two-sided boost: Q_min uses max(min(Q_one, Q_two), max(Q_one/c, Q_two/c)) where c=3.0
• Extreme midpoints (<0.10 or >0.90): strict min(Q_one, Q_two) — no c=3.0 safety net
• Q chain: Q_one -> Q_two -> Q_min -> Q_normal -> Q_epoch -> Q_final
• Sampling: 10,080 one-minute samples per weekly epoch, minimum payout $1
• Distribution: Daily at midnight UTC

Reward Pools (April 2026)

Sport	Pool/Game	Notes
Champions League QF	$24,000	Highest pool, 3-day window
EPL	$10,000	Weekend-heavy
NBA	$7,700	Frequent events, high adverse selection
CS2 A-Tier	$5,500	Lighter competition
IPL Cricket	$4,500	Emerging market
UFC Main Card	$4,250	Short event window
MLB	$1,650	Low pool, skip unless competition minimal
NHL	$1,500	Low pool

Infrastructure

• CLOB REST: https://clob.polymarket.com
• Market WS: wss://ws-subscriptions-clob.polymarket.com/ws/market
• User WS: wss://ws-subscriptions-clob.polymarket.com/ws/user
• Sports WS: wss://sports-api.polymarket.com/ws
• RTDS WS: wss://ws-live-data.polymarket.com
• Auth: EIP-712 (L1) -> HMAC-SHA256 (L2), Signature Type 2 (GNOSIS_SAFE)
• Fees: Makers 0%, takers 1.5% (crypto) / 0.3% (sports). Fee formula: fee = C * feeRate * p * (1-p)
• Heartbeat: 10-second window with 5-second buffer. Miss = ALL orders cancelled.
• Rate limits: POST /order 3,500/10s burst; POST /orders (batch, 15/req) 1,000/10s = 15,000 effective; DELETE /order 3,000/10s; DELETE /cancel-all 250/10s
• Tuesday restart: 7:00 AM ET, ~90s downtime, HTTP 425 (Too Early)
• Post-Only orders: Guaranteed maker status (0% fees), rejected if would cross spread
• GTD minimum: 60 seconds (expiration = now + 60 + N)

Smart Contracts (Polygon)

Contract	Address
CTF Exchange	0x4bFb41d5B3570DeFd03C39a9A4D8dE6Bd8B8982E
Neg Risk CTF Exchange	0xC5d563A36AE78145C45a50134d48A1215220f80a
Neg Risk Adapter	0xd91E80cF2E7be2e162c6513ceD06f1dD0dA35296
Conditional Tokens (CTF)	0x4D97DCd97eC945f40cF65F87097ACe5EA0476045
USDC.e	0x2791Bca1f2de4661ED88A30C99A7a9449Aa84174
UMA Adapter	0x6A9D222616C90FcA5754cd1333cFD9b7fb6a4F74
UMA Optimistic Oracle	0xCB1822859cEF82Cd2Eb4E6276C7916e692995130
Gnosis Safe Factory	0xaacfeea03eb1561c4e67d661e40682bd20e3541b
Polymarket Proxy Factory	0xaB45c5A4B0c941a2F231C04C3f49182e1A254052

Risk Management Defaults (ImMike config)

• maker_fee_bps: 0 / taker_fee_bps: 150
• estimated_gas_per_order: 0.02 (~$0.02 on Polygon)
• slippage_tolerance: 0.02 (2% for arb, use 1% for market making)
• max_retries: 3 / retry_delay_seconds: 1
• heartbeat_interval: 30 (bot internal, NOT Polymarket heartbeat)
• mm_enabled: false — "markets too efficient"

Genius Context

Full extraction intelligence: genius.md — 21 patterns, 13 hidden knowledge items, 7 signature moves, quality rubric, worked examples, API details

Execution Prompts (structure-pure v2)

3 deterministic practitioner prompts — each carries an Output Contract, Output Skeleton, and Quality Gate. When a deliverable matches one, Read it and honor its contract instead of improvising the output shape.

• Polymarket Market Maker — Adverse Selection Defense Configuration — skills/prediction-market-making/references/prompts-v2/adverse-selection-defense-config.md
• Polymarket Market Maker — Market Selection & Deployment Plan — skills/prediction-market-making/references/prompts-v2/market-selection-deployment-plan.md
• Polymarket Market Maker — Reward Optimization Tuning Report — skills/prediction-market-making/references/prompts-v2/reward-optimization-tuning-report.md