A Longitudinal Panel of GitHub Engineering Velocity for Venture-Backed Startups: Dataset and Early Observations

In the high-stakes world of venture capital, where investors bet on the future of new companies, the traditional method of finding the next big thing has long relied on personal networks. Investors often wait for a founder to make an introduction or for a company to announce it is raising money. However, a quiet shift is occurring. Instead of waiting for a phone call, many investors are now looking for digital footprints left behind by the companies themselves. Before a startup ever speaks to a bank or a potential investor, its engineering team is often already at work, writing code and building software. This work happens in public spaces on the internet, specifically on a platform called GitHub, where developers share their projects. The core idea is simple: the speed and volume of this public work can serve as an early warning system, revealing which companies are growing fast and which are struggling, often before any official news is released.

A new study by Maryan Kindrat takes this concept and turns it into a concrete, measurable record. The researcher has compiled a detailed log of engineering activity for 55 venture-backed startups across 20 different industries, ranging from artificial intelligence to gaming and legal technology. This log covers five quarters of activity, from the second quarter of 2025 through the second quarter of 2026. For every company in the group, the study tracks specific actions over rolling two-week periods: the commit velocity (the rate of code updates), the count of unique people who contributed to the project, and whether the company created new digital repositories, which are essentially folders where software projects live. By watching these numbers over time, the researcher can spot distinct patterns that signal what a company is actually doing under the hood.

The study identifies four main types of activity that suggest a company is accelerating. The most common pattern, appearing in three out of every four observations, is called a framework migration. This happens when a team makes a concentrated burst of changes to the underlying rules and configurations of their software without adding new people to the team. It often signals that the company is rewriting its product to be faster or more scalable. A second pattern is an engineering hiring burst, where the number of people contributing to the code grows faster than the amount of code written, suggesting the team is expanding rapidly. A third pattern, infrastructure buildout, occurs when a company starts creating new public project folders, often indicating they are building a new platform or preparing for a major enterprise launch. The final pattern is a deploy frequency spike, where the speed of coding increases sharply without a proportional increase in staff, which usually means a small team is sprinting to finish a specific milestone.

When the researcher analyzed the data from these 55 companies, the results revealed a landscape of intense and varied activity. Across the 219 recorded periods, the typical company made about 71 updates to its code in a two-week window, though the average was pulled higher by a few extremely active projects that reached nearly 400 updates. The pace of change was volatile; from one quarter to the next, some companies slowed their work by nearly half, while others accelerated their output by more than 1,600 percent. Nearly half of all the observations showed positive growth in coding speed. The study also noted that the type of activity varied by location. In the United States, the data tended to show more hiring bursts and rapid deployment spikes, whereas companies in Europe were more likely to show signs of framework migration. However, the study is careful to note that these geographic differences are preliminary, as the data for many companies could not be pinned to a specific country.

Despite the clarity of these patterns, the study is explicit about what it cannot yet tell us. The data describes the left side of the equation: the engineering activity. It does not yet link these activities to the right side: the actual financial outcomes. The researcher has not yet connected these coding patterns to specific funding announcements, stock market listings, or acquisitions. This is a deliberate limitation of the current release. The goal is to provide a clean, public baseline of engineering signals that other researchers can use. By joining this data with external records of when companies raised money, future studies could determine if a spike in coding activity truly predicts a successful investment. For now, the work stands as a detailed map of the digital construction sites where the next generation of companies is being built, offering a rare, objective look at the rhythm of innovation before the headlines ever arrive.