Lazy108/synth-price-data

Synth Price Data

Historical price data for the Bittensor Synth subnet (subnet 50), formatted as realized price paths matching the exact validator scoring intervals.

191,447 price path windows across 12 assets and 2 horizons.

Dataset Files

File	Horizon	Time Increment	Points/Window	Assets	Entries
synth_dataset_24h.jsonl	24h	300s (5 min)	289	12	121,633
synth_dataset_1h.jsonl	1h	60s (1 min)	61	5	69,814

Format

Each line is a JSON object:

{
  "asset": "BTC",
  "horizon": "24h",
  "start_time": "2024-06-05T00:00:00Z",
  "time_increment": 300,
  "time_length": 86400,
  "real_prices": [67869.22, 67798.38, ...],
  "source": "historical_1m"
}

Field	Type	Description
asset	string	Subnet asset name (BTC, ETH, SOL, etc.)
horizon	string	"24h" or "1h"
start_time	string	ISO 8601 UTC start time
time_increment	int	Seconds between price points (300 for 24h, 60 for 1h)
time_length	int	Total window in seconds (86400 for 24h, 3600 for 1h)
real_prices	list[float]	Realized prices at each interval (289 or 61 points)
source	string	Data source identifier

Asset Coverage

24h Dataset (12 assets, 5-min intervals)

Asset	Class	Source	Windows	Coverage
BTC	Crypto	Binance	17,496	2 years
ETH	Crypto	Binance	17,496	2 years
SOL	Crypto	Binance	17,496	2 years
XRP	Crypto	Binance	17,496	2 years
HYPE	Crypto	Bybit	7,878	~11 months
SPYX	Stock	Polygon.io	7,369	~2 years
NVDAX	Stock	Polygon.io	6,683	~2 years
TSLAX	Stock	Polygon.io	6,819	~2 years
AAPLX	Stock	Polygon.io	7,656	~2 years
GOOGLX	Stock	Polygon.io	7,475	~2 years
XAU	Commodity	Polygon.io	6,676	~2 years
WTIOIL	Commodity	yfinance	1,093	~71 days

1h Dataset (5 assets, 1-min intervals)

Asset	Class	Source	Windows	Coverage
BTC	Crypto	Binance	17,536	2 years
ETH	Crypto	Binance	17,536	2 years
SOL	Crypto	Binance	17,536	2 years
XAU	Commodity	Polygon.io	11,723	~2 years
HYPE	Crypto	Bybit	5,483	~8 months

Data Sources

• Binance API: BTC, ETH, SOL, XRP — 1-minute and 5-minute klines (2 years)
• Bybit API: HYPE — 1-minute and 5-minute klines (~11 months, all available)
• Polygon.io: SPYX, NVDAX, TSLAX, AAPLX, GOOGLX, XAU — 1-minute and 5-minute aggregates (~2 years)
• yfinance: WTIOIL — 5-minute data (~71 days, limited by source)

Usage

import json

# Load 24h dataset
with open("synth_dataset_24h.jsonl") as f:
    for line in f:
        entry = json.loads(line)
        # entry["real_prices"] has 289 price points at 5-min intervals
        # Use with tune_walk_forward.py, purged_walkforward_backtest.py, etc.
        break

Notes

• Stock tickers map to subnet names: SPY→SPYX, NVDA→NVDAX, TSLA→TSLAX, AAPL→AAPLX, GOOGL→GOOGLX
• Commodity tickers: GC=F→XAU (Gold), CL=F→WTIOIL (Crude Oil)
• All timestamps in UTC
• Prices are from exchange APIs (Binance/Bybit/Polygon/yfinance), not Pyth Network
• WTIOIL 5m data limited to ~71 days — WTIOIL only receives 24h prompts on the subnet
• HYPE data starts from coin launch (~July 2025) — all available history included