Same-same

Tl;dr

Cross-venue NBA game markets on Polymarket and Kalshi and generate GO signals when a paired YES/NO position locks a positive edge after fees and execution constraints. Success requires reliable schedule coverage, accurate team/market mappings, timely price polling with escalation, and a simulation path to validate fees and slippage before live execution.

Success criteria checklist (template)

✅ Market exists: Found market: <EVENT_OR_MARKET_ID>
✅ Market is active: Status: <status>
✅ Market not closed: Closes: <ISO_8601>
✅ Valid side: Side: <yes|no>
✅ Order size >= 1 contract(s): Size: <n> contract(s)
✅ Order notional >= $1.00: Order value: <$n.nn> (min: $1.00)
✅ Price in valid range (0-100¢): Price: <n>¢
✅ Market has reasonable spread (<50¢): Spread: <n>¢ (Bid: <n>¢, Ask: <n>¢)
❌ Market has real liquidity (not placeholder prices): Bid: <n>¢, Ask: <n>¢ - appears to be placeholder prices
❌ Price is fillable (>= bid: <n>¢): Your price: <n>¢
✅ Sufficient account balance: Balance: <$n.nn>, Cost: <$n.nn>

Opportunity definition

• Instruments: NBA game winner markets on Polymarket (binary) and Kalshi (binary). We pair a YES on one venue with a NO on the other.
• Edge: total cost of paired legs + fees below $1.00 (100 cents), with fillability and liquidity checks.
• Output: GO/NO-GO signal per game + venue pair with justification and expected edge.

Inputs and data sources

• NBA schedule: nba-2025-PacificStandardTime.csv (current baseline).
• Polymarket: market metadata and CLOB top-of-book prices (source to be confirmed).
• Kalshi: market list and quotes (source to be confirmed).

Schedule ingest

• Lookahead window: default 48 hours (configurable).
• Flow: read all games within the window, then derive candidate market identifiers/URLs for each venue.
• Uncertain timing: game time alignment rules are still open; must account for venue-specific market close times.

Mapping table (teams)

Store a mapping table to align team identifiers across venues and internal references.

Fields:

• internal_slug (example: nba-la-lakers)
• common_name (example: L.A. Lakers)
• polymarket_team_id
• polymarket_team_name
• kalshi_team_id
• kalshi_team_name
• source_name (where the mapping was obtained)
• source_key (identifier from the source)
• last_verified_at

Market matching (game level)

• Build a canonical game id from schedule: date, home_team, away_team.
• Generate candidate venue markets using team mapping and venue naming patterns.
• Keep a one-to-many view until we can confirm the true venue market id for the game.

Pricing and polling

• Baseline polling: every 60 seconds for all candidate games.
• Escalation: switch to 10-second polling after volume threshold is met (configurable).
• Persist every snapshot to the DB with venue, market id, bid/ask, size, and timestamp.

GO signal logic (draft)

• Pair legs: example Poly YES (Team A) + Kalshi NO (Team A).
• Compute edge_cents = 100 - (poly_leg_cost + kalshi_leg_cost + fees_cents).
• Require:
  • market exists and active
  • market not closed
  • valid side
  • size and notional minimums
  • price in valid range
  • spread below threshold
  • real liquidity (not placeholder prices)
  • price is fillable
  • account balance sufficient
• If all pass and edge_cents >= min_edge_cents, emit GO.

Simulator

• Run the same signal logic with configurable inputs:
  • fees by venue
  • min order size and notional
  • slippage model
  • rate limits and polling cadence
  • fill probability threshold

Open questions

• How to normalize market close times relative to scheduled tipoff.
• The canonical source for schedule (CSV is placeholder vs. future API feed).
• Polymarket and Kalshi naming conventions for NBA markets (slug/ticker patterns).
• Venue fee models and minimums for NBA markets.

Implementation references

• app/agents/ (existing opportunity work; review for compatible patterns)
• docs/opportunities/time-alignment-approx.md (format reference)