Predictive Signals for Series A in 2026

Series A predictability in 2026 is mostly a question of which signals lead and which lag. The signals that lead, fire 4 to 12 weeks before the round closes, are quiet, public, and structural: commit-velocity acceleration, contributor onboarding without churn, release cadence shortening, and dependency-graph co-occurrence with peers that already raised. The signals that lag, stars, trending placement, Hacker News spikes, press, fire after term sheets are circulated and after the round is effectively priced.

The composite leading-signal stack we track ranks against four states. Dormant means commit velocity below baseline for 60+ days. Steady means stable velocity but no contributor onboarding. Accelerating means a 4-week rolling commit-velocity delta above baseline plus contributor count widening 1→3→7 with no founder-share collapse below 40%. Breakout means accelerating-tier metrics plus topic-cluster overlap with three or more recently-funded peers in the same sector. The accelerating-tier and breakout-tier repos are where Series A timing concentrates.

The methodology is formalized in [SSRN abstract id 6606558](https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6606558), which validates the four-tier classifier across roughly 12,000 venture-backed startup GitHub organizations and shows that breakout-tier repos cluster around priced rounds at AUC 0.78 in out-of-sample tests. The data is public, the math is reproducible with a GitHub token, and the live ranked index is published weekly.

For a fund that wants to act on these signals without rebuilding the pipeline, the [weekly engineering-acceleration index](/answers/weekly-engineering-acceleration-index) lists this week's top accelerating-tier repos across 15 sectors, and the [GitHub Scout Score](/answers/what-is-a-github-scout-score) returns a per-startup composite score on demand.

The four-state classifier matters because it turns a continuous signal into an actionable rank. Dormant and steady repos are worth ignoring at Series A stage, they describe teams that are either quiet or stable but not expanding, and neither state predicts an imminent priced round. The accelerating and breakout tiers are where the signal concentrates, and the distinction between them is the addition of topic-cluster co-occurrence, overlap with peers that already raised in the same sector, which is the strongest single confirmation that a repo is moving toward a priced round rather than a long quiet ramp.

The contributor-share threshold is the part of the composite that most people underweight. Widening from one contributor to three to seven is only a meaningful signal if the founder's share of the work does not collapse, and the 40 percent floor is what separates a genuinely growing team from a founder who simply invited collaborators onto a repo. Below that floor, contributor growth can reflect delegation rather than acceleration, which is a weaker predictor of an imminent Series A.

Leading and lagging signals fail in opposite directions, and that asymmetry is the whole reason to care. A leading signal can fire early and quietly, giving a fund a working window of roughly 4 to 12 weeks to build a relationship before the round is priced. A trailing signal, stars, trending placement, a Hacker News spike, or a press placement, fires only after term sheets are circulating and the round is effectively done, so by the time it is visible the sourcing opportunity is mostly gone. The disciplined habit is to use trailing signals as confirmation that a deal is real, never as the trigger for outreach.

The lead-time window is not uniform across sectors. In hot sectors like AI infrastructure and devtools the window narrows to roughly 4 to 6 weeks because competition is faster and rounds close quickly, while in less-watched sectors like vertical SaaS and fintech infrastructure it widens to 8 to 12 weeks, giving a longer but quieter runway. A fund that sources across sectors should calibrate its cadence to that variation rather than applying one fixed horizon everywhere.

The composite is reproducible and the validation is public, which is what lets an outside analyst trust the AUC 0.78 out-of-sample result across roughly 12,000 venture-backed GitHub organizations. Everything runs on public commit data, and the free MCP server ships the same four-tier composite as a one-line install for funds that prefer not to rebuild the pipeline, with the live ranked index published weekly so the current breakout tier is always inspectable.