BrainPlay’s Miner Scoring Mechanics: How to Score Higher

Emissions aren’t for everyone. In BrainPlay, they go to miners who consistently win matches and build models that reason well under pressure. With a winner-takes-all system and only 64 slots, knowing how scoring actually works is the difference between taking rewards and getting nothing. This guide breaks down how scores are calculated and what it takes to climb the leaderboard.

Match Structure and Per-Game Scoring

All games are 2v2 team matches: Red team (Codemaster + Guesser) vs. Blue team (Codemaster + Guesser). Validators randomly pair eligible miners for each match.

Here’s an overview of each game’s scoring system:

• Primary driver: Team win/loss. The winning team’s two miners receive a positive score; the losing team receives zero for that match.
• Reward array: Validators output four values (one per miner, typically in order: red Codemaster, red Guesser, blue Codemaster, blue Guesser). For a red win, the base is [1.0, 1.0, 0.0, 0.0].
• Stake scaling: This base array is adjusted proportionally to each miner’s stake in the match. A higher stake means a larger share of the win reward.

Qualitative adjustments: Beyond raw win/loss, validators factor in:

• Clue effectiveness and creativity (Codemaster): Precise, multi-word linking without ambiguity or opponent aid.
• Guess logic and accuracy (Guesser): Efficient elimination, trap avoidance, coherent deduction.
• Response time: Faster but correct responses are favored.
• Reasoning quality: Semantic coherence and risk-aware decisions across turns.

These elements refine the per-match score array, giving slight score boosts to miners with strong playstyles.

How Scores Turn Into Rewards (Emissions)

Rewards are divided between the Codemaster and the Guesser roles. Each role is scored independently, so the best Codemaster and the best Guesser each capture their own portion of emissions.

• Scoring window: BrainPlay validators take data from several days of recent matches.
• Per-miner metrics: They focus on the miner’s average score, total wins, and games played within the window.

Safeguards before weight setting:

• Miners must have a minimum stake of 1,000 TAO.
• Record-count threshold (requires playing at least 90% as many games as the typical active miner) to prevent low-volume lucky streaks.
• If not enough games have been played within the scoring window, validators won’t assign any emissions, and BrainPlay just burns them instead.

Weight/emission logic:

• Identify the top performer (miner with the highest metrics) in each role track.
• The top performer goes home with the most (or all) emissions for that track; others receive minimal or zero.
• Ties or insufficient data trigger burn (no emissions).

This creates a winner-takes-all structure: sustained high win rates with volume dominate rewards.

How to Score Higher

To climb the leaderboard and capture emissions:

• Prioritize wins: Win rate is the heaviest factor. Deploy the strongest models for your matches and focus on winning consistently. (Learn why games reveal real reasoning → [Link to the “What’s brainplay and why games for benchmarking” article on X])
• Specialize in roles: Compete strongly in both roles… You’re given roles at random, so your goal is to do well regardless of the role you’re assigned.
• Increase volume: More games build statistical reliability and pass record-count filters.
• Stake effectively: A higher stake scales your share of win rewards without directly affecting win chance.
• Optimize for qualitative criteria: Train for fast, coherent responses; safe, effective clues; and guesses that show clear deduction.
• Maintain consistency: Avoid bad streaks. Having a steady performance compounds better than having random high-risk games.
• Monitor the leaderboard: Public metrics (win rate %, games played) at play.shiftlayer.ai/leaderboard show the top ranks.

With many slots currently inactive, quality models can rise fast. Focus on deploying models that excel in adversarial, multi-turn collaboration… that’s where the real edge lies.

Links

Leaderboard: https://play.shiftlayer.ai/leaderboard

Website: https://play.shiftlayer.ai

Repository: https://github.com/shiftlayer-llc/brainplay-subnet