PHBench

Predict the next
Series A
from a 24h launch.

PHBench is an open benchmark for predicting Series A funding from Product Hunt launch signals. Models are trained and ranked on 67,292 launches across seven years. Our best model scores every new launch.

RESEARCH BY

The base rate

0.78% of launches
raise a Series A.

A needle in a haystack. Only 0.78% of launches raise a Series A. Our best model achieves 4.7× lift over random on the held-out test set.

66,764 launches → no Series A528 → Series A within 18mo

LAUNCHES

67,292

2019 → 2025

WINNERS

528

verified Series A

WINDOW

18 mo

launch → term sheet

PERIOD

7 yr

train / val / test

LAUNCHING ON PRODUCT HUNT?All tools →

Series A checker →

Paste your launch link to see where our model ranked it for Series A that week.

Upvote calculator →

How many upvotes it takes to finish #1, top 3, top 5 or top 10, by launch day.

Best time to launch →

The best launch day for your category, and launch time in your time zone.

Rank predictor →

Predict your launch-day rank from your hunter, team, day and media. No sign-up.

01THE LEADERBOARD

code

dataset

paper

FILTER

#▴	MODEL	ORGANIZATION	METHOD	F0.5	AP	REC	AUC
01	Top-3 Ensemble🏆	AUTHOR	avg(ENS_avg, ENS_ISO, XGB_m21)	★0.284	0.126	0.080	0.840
02	ENS_avg🥈	AUTHOR	Ensemble average	★0.271	0.094	0.080	0.836
03	LGBM FS3 + ISO🥉	AUTHOR	LightGBM + isotonic calibration	★0.247	0.066	0.080	0.793
04	XGB FS4 + ISO	AUTHOR	XGBoost + isotonic calibration	★0.237	0.074	0.100	0.845
05	XGB FS4	AUTHOR	XGBoost	★0.223	0.070	0.100	0.829
06	Gemini 3 Flash	Google	LLM — zero-shot, anonymized signals	★0.129	0.034	0.060	0.713
07	LR FS4🏁BASELINE	AUTHOR	Logistic Regression	0.127	0.047	0.060	0.825
08	Gemini 2.5 Flash	Google	LLM — zero-shot, anonymized signals	0.067	0.022	0.060	0.697
09	Gemini 3.1 Pro	Google	LLM — zero-shot, anonymized signals	0.057	0.023	0.050	0.725

Showing 9 of 9 models · val set · ★ beats LR baseline

02SIGNALS · WHAT THE DATA SAYS

Six features that predict a Series A — and four that don't.

61 FEATURES ENGINEERED. 12 SIGNAL. 4 NOISE.

XGBOOST GAIN IMPORTANCE × SERIES A LIFTn=67,292 · XGBoost gain importance

Daily rank on launch

LIFT

3.50×

IMPORTANCE

0.85

p-VALUE

< 0.01

DISTRIBUTION · WINNERS vs LOSERS

LOSERSWINNERS

FINDING

Top-3 products on launch day raise Series A at 3.5× the base rate. rank_bucket captures this non-linearly — moving from #4 to #1 adds more lift than #10 to #4.

03METHODOLOGY

Reproducible. Citable.

We built PHBench the way we'd want to be evaluated. Every label is manually audited. Every feature is documented. Every submission reruns on a hash-pinned test set.

The test set phbench_public_test.csv is held out until final submission.

CITE THIS WORK

@article{ihlamur2026phbench,
  title={PHBench: A Benchmark for Predicting Startup Series A
    Funding from Product Hunt Launch Signals},
  author={Ihlamur, Yagiz and Griffin, Ben and Chen, Rick},
  journal={arXiv preprint arXiv:2605.02974},
  year={2026},
  url={https://arxiv.org/abs/2605.02974}
}

04BENCHMARK INFORMATION

Benchmark Information

Predict the nextSeries Afrom a 24h launch.

Predict the next
Series A
from a 24h launch.