Which GitHub Metrics Predict Startup Fundraising?

No GitHub metric predicts fundraising with certainty; three of them correlate with pre-announcement behavior reliably enough to be worth watching. The correlation runs through a mechanism, not magic: rounds fund scaling, and scaling leaves engineering exhaust before it leaves a press release.

The three metrics with a mechanism. Commit-velocity change: acceleration in weekly commits reflects shipping intensity rising, typically as a team gears a product toward the milestone a raise will fund. Distinct-contributor growth: new engineers appear on public repos weeks before headcount is announced, hiring is the single most direct pre-raise tell. Repository and dependency expansion: new repositories, languages, and dependencies map to scope growth, second products, infrastructure builds, integrations that suggest a company expanding its surface area. Each is observable, timestamped, and hard to backdate.

The measured lead times, with their sample. Across the tracked set (350+ venture-relevant organizations, methodology published), engineering acceleration precedes public fundraise announcements by 3-6 weeks on average, and precedes database coverage by 6-12. Those are distributions, not promises: some rounds follow acceleration by a quarter, some never follow. The correct use is priors for attention, not predictions for allocation.

What does not work. Raw star counts: popularity lags substance and is campaignable. Total commits all-time: a stock measure, blind to change. Single-week spikes: releases and rebases masquerade as acceleration. Any single metric alone: the false-positive rate only becomes acceptable when velocity, contributors, and repositories move together, which is exactly the confirmation rule this site's signal layer applies.

The honest ceiling. These metrics see technical teams only, and see execution, not outcomes: nothing in commit history tells you revenue, retention, or founder quality. They are a when-to-look layer over the judgment stack you already run. Used that way, the cost is zero (the feed and the MCP server are free) and the edge is real: you meet companies during the 3-6 week window when engineering is visibly scaling and the round is still private. The full definitions, normalizations, and per-metric track record are on the methodology page.

The signals combine into one confirmation rule. Individually, commit velocity, contributor growth, infrastructure-shaped commits, and repository-creation bursts are each noisy, and any single one is easily fooled. The false-positive rate only becomes acceptable when several move together, which is exactly the confirmation rule this site's signal layer applies before flagging a name. Watch for velocity sustained across multiple 14-day windows, contributor growth alongside it, and a burst of new repositories that look like auth, CI, or observability work.

Reproducibility is built in, not claimed. The validation set of 219 startup-period observations is published in the SSRN preprint, the raw dataset is archived on Zenodo under a Creative Commons license, and the classifier itself is open source, so anyone can rerun the analysis or extend it. That is the difference between a metric you can argue about and a metric you can audit.

What private development does to the signal. A startup that does most of its work in private repositories will be under-represented in commit and contributor counts. The methodology weights the public signal against the org's total public footprint to compensate, but it cannot recover signal from genuinely private development, so treat an absent or thin signal as unknown, not as negative.

Stars are not the same thing. A repository can spike in stars from a single Hacker News post without any team expansion or shipping acceleration, because stars measure attention while commit velocity and contributor growth measure sustained engineering investment. The latter is what actually predicts a fundraise.

The right way to use the output. These metrics see execution, not outcomes; nothing in commit history tells you revenue, retention, or founder quality. Use them as a when-to-look layer that surfaces candidates during the 3-6 week window before a round is announced, then run your normal diligence on top.