WC Knockout — FunctionSpace

By FunctionSpace — built with Simmer's agent trading infrastructure. Play at propSPACE — signup at app.propspace.fun. Play-money only. Real USDC prizes administered by FunctionSpace.

Trade World Cup 2026 knockout-stage player fantasy-score markets. Each knockout round is a clean slate with a fresh $1,000 propSPACE play-money bankroll. The live campaign prize pool is $6,500 total: $1,000 for each knockout round from R32 through Semis, plus a $2,500 grand prize across all rounds. The skill auto-detects the current open round by default rather than targeting a stale R32 kickoff date.

🧪 Dry-run by default. Pass --live to place propSPACE play-money positions. This is a mutating competition-engine action, but it is not Simmer $SIM, Polymarket, Kalshi, or real-money trading.

How It Works

propSPACE markets are continuous scalar belief markets — the outcome is a player's aggregate fantasy score for the match, not a binary yes/no. You trade a probability distribution over the score range. The engine uses a scoring-rule mechanism: positions pay out more when the consensus shifts toward your belief.

Competition format: each round you pick 1 FWD + 1 MID + 1 DEF. Those 3 selections are locked until round settlement. The skill ranks all open markets per position by edge (your belief mean vs consensus mean) and buys the best player in each slot.

Each run:

1. Authenticates with propSPACE (passwordless — auto-creates account on first run)
2. Discovers open WC knockout player markets (~100 per round)
3. For each market: matches the player, reads FunctionSpace expected fantasy points, and applies optional sentiment adjustment
4. Builds a multimodal density recipe with a low appearance/no-return cluster and a higher return cluster
5. Groups markets by position (FWD / MID / DEF from market metadata)
6. Picks the highest-edge player per position (best 1 FWD + 1 MID + 1 DEF)
7. Buys those 3 play-money market positions when --live is passed

Strategy

Expected fantasy score = FunctionSpace expectedPts / line × optional sentiment adjustment.

The default belief recipe is now multimodal (position_type: "density"): a low appearance/no-return cluster plus a higher attacking/return cluster. This better matches propSPACE player-market pricing than a single symmetric Gaussian. Set FS_RECIPE_SHAPE=normal to use the older unimodal Gaussian baseline for comparison. Expert sentiment from WC 2026 fantasy articles can shift the mean ±15% when data/player_data.json exists and has at least 3 sentiment sources for that player.

Best pick = highest |our_belief_mean - consensus_mean| within the position slot.

⚠️ Distribution model is a baseline. FunctionSpace confirmed the engine prices player markets as multimodal distributions. The skill now submits a simple two-cluster density recipe, but FunctionSpace should still review the cluster weights/spreads against the official scoring model.

Engine URL confirmed. FS_BASE_URL defaults to the World Cup competition engine confirmed by FunctionSpace: https://fs-engine-api-mech-v0-4.onrender.com.

Setup

1. Set your username and engine URL

   export FS_USERNAME=simmer_agent_1
   export FS_BASE_URL=<competition engine URL from FunctionSpace>
   

2. Enrich sentiment (optional but improves pick quality)

   export BRAVE_API_KEY=your_key
   python3 scripts/enrich_from_web.py --round "current open round"
   

3. Dry run — see which 3 players the skill would pick

   python3 main.py
   

4. Browse markets to see what's live with position labels

   python3 main.py --list-markets
   

5. Inspect a specific market

   python3 main.py --inspect 261
   

6. Go live — executes 1 FWD + 1 MID + 1 DEF propSPACE play-money trade

   python3 main.py --live
   

Quick Commands

python3 main.py                    # dry run: pick best FWD/MID/DEF
python3 main.py --live             # execute 3 picks
python3 main.py --list-markets     # list open markets with position labels
python3 main.py --market 261       # target one market
python3 main.py --inspect 261      # show consensus vs belief
python3 main.py --all-markets      # trade all markets above edge threshold

Configuration

Player Data

The current production path uses each market's FunctionSpace metadata.expectedPts / line as the base expected fantasy score and then optionally applies sentiment enrichment. If data/player_data.json exists, it can add player aliases and sentiment fields; if it is absent, markets still run using the FS line as-is and print player not in DB — using FS line as-is.

fs_beliefs.py is retained as a legacy reference for the older raw-vector convention (num_buckets + 2) and is not imported by main.py.

Before Going Live — Required from FunctionSpace

1. Scoring/variance confirmation — the base expected score now comes from FunctionSpace market metadata (expectedPts / line), while position-aware spread uses the constants in main.py. Confirm those spread assumptions against the official scoring model.
2. Competition engine URL — update FS_BASE_URL if FunctionSpace moves from the mech-v0-4 engine.
3. Market metadata schema — confirm the field name for player position in market metadata (checked as position, positionType, player_position in that order). Correct it in _market_position() if needed.
4. Access code — if signup requires a partner code, set FS_ACCESS_CODE.
5. Market range / density review — lower_bound, upper_bound, num_buckets, and the multimodal density recipe should be reviewed against the current round's market ranges with --inspect <id>.
6. Competition backend / rules confirmation — confirm with FunctionSpace that https://fs-engine-api-mech-v0-4.onrender.com is the same backend used to settle the live propspace.fun competition, that API trades count toward the $6,500 campaign, and that agent/bot trading is allowed under the competition rules.