Six months ago I started tracking the GitHub activity of 350+ startups across 15 sectors. The goal: find public signals that predict which startups are about to raise.
I backtested against 219 actual fundraises. Here's what I found.
The setup
Every week, my system pulls data from the GitHub API for 350+ startup organizations:
- Commit counts per repository
- Active contributor counts
- New repository creation
- Infrastructure file changes (CI/CD, Dockerfiles, monitoring configs)
I normalize by sector (a Healthcare startup ships differently than a Dev Tools startup) and score each startup on an Engineering Acceleration Score.
Live data is at signals.gitdealflow.com.
Three signals that predict fundraises
1. Commit velocity acceleration (the strongest signal)
Startups that raised within 90 days showed a median 34% week-over-week increase in commit velocity. The pattern is distinctive: steady baseline for months, then a sharp ramp.
This makes intuitive sense. Teams push hard to ship milestones before pitching investors. The code acceleration starts before the deck is finished.
2. Contributor count spike
Fundraising startups added a median of 2.3 new active contributors in the 60 days before their round. This reflects pre-round hiring pushes. You're building the team before you ask for money.
3. Infrastructure buildout
Startups about to raise were 3.2x more likely to add new CI/CD workflows, Dockerfiles, or monitoring configs in the 90 days pre-round. This signals preparing for scale (and investor demos).
Sector benchmarks (Q2 2026)
| Sector | Startups | Median weekly commits | Top quartile |
|---|---|---|---|
| AI & ML | 67 | 47 | 112 |
| Developer Tools | 34 | 89 | 203 |
| Fintech | 42 | 34 | 78 |
| Web3 | 28 | 56 | 134 |
| Healthcare | 19 | 22 | 51 |
| Enterprise SaaS | 45 | 38 | 84 |
Full table across all 15 sectors at signals.gitdealflow.com.
The honest result
My first pre-registered cohort went 0-for-10. Zero correct predictions out of 10 startups I said would raise within 90 days.
I published that result. Then I refined the methodology, expanded the dataset, and re-ran the backtest against 219 actual fundraises.
The improved signals now identify acceleration patterns that precede fundraises. But I lead with the failure because that's what makes the data trustworthy.
Full paper with the 0-for-10 transparency ledger: SSRN abstract 6606558.
How to use this
If you're an angel or VC:
- Track GitHub activity for startups in your thesis sectors
- Watch for the three-signal composite (velocity + contributors + infra)
- When all three fire, the startup is likely 30-47 days from announcing a round
- Reach out before the round is announced
The automated version of this is at GitDealFlow. You can also browse the open data at signals.gitdealflow.com.
Open source everything
- Methodology: SSRN paper
- Due diligence checklist: GitHub
- Live data: signals.gitdealflow.com
- Signal interpretation guide: GitHub
The author runs GitDealFlow. All methodology is open. CC BY 4.0.
Top comments (0)