YC Bench: a Live Benchmark for Forecasting Startup Outperformance in Y Combinator Batches

Paper: https://arxiv.org/abs/2604.02378

Forecasting which startups will dominate a YC batch — months before Demo Day.

---

Overview

YC Bench turns every Y Combinator batch into a rapid-evaluation environment for startup success prediction. Instead of waiting 7–10 years for exits or large funding rounds, we use the Pre-Demo Day Score — a short-term proxy metric that combines public traction signals and Google web mentions.

This repository contains all data, collection scripts, scoring code, and analysis for the YC W26 batch (196 startups).

Repository Structure

ycbench/
├── yc_w26_startups.csv              # Main list of YC W26 companies
├── yc_mentions.csv                  # Google mentions during the batch
├── yc_mentions_early.csv            # Pre-application Google mentions (baseline)
├── YC_W26_Google_Mentions.ipynb     # Colab notebook for data collection & visualizations
├── yc_google.py                     # Google mentions scraping utilities
├── requirements.txt
├── scripts/
│   ├── scrape/                      # Data collection scripts
│   ├── processing/                  # Data cleaning pipelines
│   └── scoring/                     # Pre-Demo Day Score computation
├── fix_pipeline.sh
├── paper/                           # LaTeX paper
└── figures/                         # Charts from the paper

Key Features

• Scripts to collect fresh Google mentions data
• Colab notebook for easy data collection and visualization
• Pre-computed mentions (during batch + pre-application baseline)
• Traction data integration
• Baseline model (pre-YC application Google mentions)

Quick Start

1. Clone the repository

git clone https://github.com/benstaf/ycbench.git
cd ycbench
pip install -r requirements.txt

2. Explore with Colab (Recommended)

Results — YC W26 Batch

A simple baseline using Google mentions before the YC application deadline achieved:

• Precision@20: 70%
• Recall@11: 55%
• Lift over random: 7×
• Forecasting horizon: ~5 months

Full details are available in the paper.

Paper

Title: YC Bench: a Live Benchmark for Forecasting Startup Outperformance in Y Combinator Batches
Author: Mostapha Benhenda

📄 https://arxiv.org/abs/2604.02378

Citation

@misc{benhenda2026ycbench,
  title={YC Bench: A Live Benchmark for Forecasting Startup Outperformance in Y Combinator Batches},
  author={Mostapha Benhenda},
  year={2026},
  url={https://arxiv.org/abs/2604.02378}
}

Roadmap

• Support for future batches (S26, W27, ...)
• Learn optimal signal weights from historical data
• Expand traction dataset
• Public leaderboard for community models

Contributing

Contributions are welcome! Especially:

• Improved scraping methods
• New predictive signals
• Better scoring logic
• Support for upcoming batches

Feel free to open issues or submit pull requests.

---

Built to make startup forecasting faster and more rigorous.
Star the repo if you're working on this problem! 🚀