---
license: mit
datasets:
- raulcel/sn28-miner-richtao-dataset
language:
- en
pretty_name: SN28 Miner Richtao Dataset
tags:
- bittensor
- tao
- s&p500
- trading
- forecasting
---

# Dataset Card for SN28 Miner Richtao Dataset

This dataset contains the real-time prediction dumps generated by the **Richtao miner** operating in the **Bittensor Subnet 28 (S&P 500 Oracle)** network.  
It is used to evaluate and benchmark intraday forecasting models for the S&P 500 index.

---

## 🧾 Dataset Details

- **Curated by:** Raúl Celis  
- **Funded by:** Private research / Bittensor TAO network  
- **Language:** English  
- **License:** MIT  
- **Repository:** https://huggingface.co/datasets/raulcel/sn28-miner-richtao-dataset  
- **Purpose:** Store public prediction logs (`preds.csv`, `truths.csv`, `metrics.json`) for model transparency and RMSE tracking.

---

## 📈 Dataset Description

Each record represents a 5-minute-interval prediction of the S&P 500 index price made by the Richtao miner.  
Predictions are produced every minute using macro-market indicators (bond yields, VIX, DXY, commodities, credit spreads, etc.) and a local news-sentiment model.

### Files
| File | Description |
|------|--------------|
| `logs/preds.csv` | Predicted S&P 500 values for +5 m to +30 m horizon. |
| `logs/truths.csv` | Realized S&P 500 values (back-filled for error metrics). |
| `logs/metrics.json` | Rolling RMSE and MAE statistics of prediction accuracy. |

---

## 🔬 Uses

### Direct Use
- Research on intraday financial forecasting.  
- Training and validation of AI models for short-term market direction.  
- Transparency and reproducibility for Bittensor Subnet 28 miners.

### Out-of-Scope Use
- Do not use for live trading decisions without independent validation.  
- Not intended for long-term forecasting or non-financial applications.

---

## ⚙️ Dataset Structure

Each row in `preds.csv` includes:  
- Timestamp (UTC)  
- SPX now and predicted values  
- ES futures levels  
- Macro changes (Δ2Y, Δ10Y, ΔDXY, ΔVIX, ΔWTI, ΔCU, ΔHYG)  
- Sentiment score (−0.5 to +0.5)  
- Prediction horizon (+5m…+30m)

---

## 🧩 Dataset Creation

### Curation Rationale
To document and quantify the intraday performance of an autonomous AI miner in Bittensor Subnet 28 (S&P 500 Oracle).

### Source Data
Aggregated from public market feeds (Yahoo Finance) and local sentiment analysis of financial headlines.

### Data Collection and Processing
Automated Python scripts running under PM2 on Ubuntu servers fetch real-time data each minute, generate predictions, and upload the logs to Hugging Face.

### Annotations
No manual annotations; all values are computed automatically from market data.

### Personal and Sensitive Information
None. All data are aggregated market metrics without personal content.

---

## ⚠️ Bias, Risks, and Limitations
Predictions are based on historical correlations and simplified rules; unexpected macro events may invalidate short-term patterns.  
Do not use the dataset for financial advice or speculative trading without risk assessment.

---

## 📚 Citation
If you use this dataset, please cite it as:

> Celis, R. (2025). SN28 Miner Richtao Dataset [v1.0]. Hugging Face Datasets. https://huggingface.co/datasets/raulcel/sn28-miner-richtao-dataset

---

## 📬 Contact
For questions or updates, contact: `raulcel (at) huggingface (dot) co`

---