Loading
Loading

Neuron registration trends
Online poker platforms struggle to identify automated play reliably. Bots and human players act within the same rules, so detection cannot depend on single actions or known signatures. A useful system must generalize across diverse players and bot families, resist identity shortcuts, and produce calibrated probabilities rather than binary labels. Static public benchmarks also invite overfitting.
Poker44 addresses this by operating a decentralized network where competing detection models are evaluated continuously against the same private ground truth. Validators lease sealed, immutable micro-session datasets derived from consented tournament play, query all eligible miners for risk scores, and score responses locally using private labels that never leave the validator. This approach prevents label leakage, ensures every model faces identical test conditions, and rewards calibrated, generalizable detectors.
On Poker44 (Bittensor subnet 126), miners operate detection models that accept ordered lists of four-decision micro-sessions and return calibrated bot-risk scores in [0,1]. Each micro-session contains only seven coarse strategic featuresβno cards, exact chip amounts, timing, IDs or labels. Validators score responses using actor-balanced metrics (average-precision, recall at 5% false-positive rate, and Brier skill) and allocate rewards deterministically to the highest-scoring miner, with the remainder distributed to a tournament-funding hotkey.
Use cases were auto-generated from the GitHub repository and may not reflect the latest changes.
Poker44 is a decentralized Bittensor subnet that evaluates poker bot detection models by having miners classify tournament micro-sessions (4-decision sequences) as human or bot-generated, using calibrated risk scores in [0,1] range. Validators query all eligible miners against sealed, immutable evaluation windows derived from consented tournament telemetry, score responses using actor-balanced average-precision, recall and Brier skill metrics, and allocate rewards to the highest-scoring miner deterministically.
Miners expose a single Axon endpoint serving MicroSessionDetectionSynapse (contract_version: microsession-v1) that accepts ordered lists of schema-v4.1 micro-sessions and returns calibrated risk_scores in [0,1] and binary predictions. Models are loaded via factory pattern at startup and inference is serialized. Validators poll the platform API (api.poker44.net) for sealed evaluation windows, lease the same immutable item ordering, verify dataset integrity, query every eligible miner concurrently, score responses locally using private labels, determine the winner deterministically, and emit weights to Bittensor consensus.
Platform enforces privacy boundary: miners receive only 7 coarse decision features per session (no cards, timing, IDs, labels, or telemetry). Validators keep labels and actor groups locally. Optional encrypted Axon commitments hide miner IP from public metagraph while allowing validator resolution.
Auto-update watcher keeps validators synchronized with main branch.
Decentralized Generative Video Platform
This information was auto-extracted from the GitHub repository and may be incomplete or out of date. Always refer to the subnet's GitHub repository for the latest instructions.
Validator scoring uses actor-balanced sample weights ensuring one prolific player cannot dominate: Quality = 0.50 * average-precision skill + 0.30 * recall at <=5% false-positive rate + 0.20 * Brier skill. Invalid/missing/non-finite/out-of-range/wrong-length responses score zero. Highest positive finite quality wins; exact ties broken by lower UID.
Scores are refreshed every 720 blocks without re-querying miners. No EMA, no legacy evaluation fallback. Dashboard reports scores/errors for observability only and does not provide weights.
The sections above were auto-generated from the GitHub repository and may not reflect the latest changes. View repository
FAQs were auto-generated from the GitHub repository and may not reflect the latest changes.
Holder & total-staked metrics as of 2h ago (hourly snapshot)
GPU optional depending on model implementation. Reference model runs on CPU. Inference serialized with process-local lock; synchronous models run outside event loop. Model loading occurs once at startup.
Requires SQLite for durable event outbox. PM2 process supervisor recommended for dual-process setup (validator + auto-update watcher). Commit-reveal weight settlement requires on-chain transaction capability.
Clone the Poker44 subnet repository
git clone https://github.com/Poker44/Poker44-subnet && cd Poker44-subnet
Install the poker44 package and dependencies from requirements.txt
pip install -e .
Configure wallet, hotkey, network UID (default Finney 126) and Axon port
export WALLET_NAME=your_wallet && export HOTKEY=your_hotkey && export NETUID=126 && export AXON_PORT=8091
Set model factory import path; defaults to reference factory if unset(optional)
export POKER44_MODEL_FACTORY=your_package.module:create_model
Configure optional model artifact path, device (cpu/cuda), and reported version string(optional)
export POKER44_MODEL_PATH=/path/to/model && export POKER44_MODEL_DEVICE=cpu && export POKER44_MODEL_VERSION=your-model-v1
Run linter and test suite before deployment
ruff check poker44 neurons tests && PYTHONPATH=. pytest -q
Start miner using PM2 process manager with configured settings
bash scripts/miner/run/run_miner.sh
Monitor miner logs for successful startup and validator requests(optional)
pm2 logs poker44-miner
Clone the Poker44 subnet repository
git clone https://github.com/Poker44/Poker44-subnet && cd Poker44-subnet
Install the poker44 package and dependencies from requirements.txt
pip install -e .
Configure wallet, hotkey, and network UID (default Finney 126)
export WALLET_NAME=your_wallet && export HOTKEY=your_hotkey && export NETUID=126
Configure emission allocation (burn, funding, winner shares); funding hotkey must be registered(optional)
export POKER44_BURN_FRACTION=1.00 && export POKER44_FUNDING_FRACTION=0.00 && export POKER44_FUNDING_HOTKEY=<your_funding_hotkey>
Configure platform API endpoint and evaluation window poll interval(optional)
export POKER44_SUBNET_DATA_URL=https://api.poker44.net && export POKER44_POLL_INTERVAL_SECONDS=300
Enable automatic validator updates from main branch (requires git repository)(optional)
export AUTO_UPDATE_ENABLED=true && export TARGET_BRANCH=main
Run linter and test suite before deployment
ruff check poker44 neurons tests && PYTHONPATH=. pytest -q
Start validator using PM2 process manager (creates poker44-validator and poker44-validator-auto-update processes)
bash scripts/validator/run/run_vali.sh
Monitor validator logs for window leases, miner queries, scoring and weight settlement(optional)
pm2 logs poker44-validator