An open prediction market on which 5 GitHub-flagged startups raises Series A first.
I shipped Series A Race 2026 at signals.gitdealflow.com/markets/series-a-race-2026. Live implied odds for 5 high-signal early-stage startups, derived from GitHub commit-velocity data. Resolves Dec 31, 2026. Public methodology, public resolver, machine-readable JSON, CC BY 4.0. No real money — we don't operate an exchange.
Key Takeaway
Polymarket and Kalshi don't list seed-stage startup markets. Too granular for real-money exchanges. So we built our own seeded version on top of the GitHub commit-velocity dataset we already maintain. Five candidates (Zapply Jobs 46% / Kanvas 22% / AtroCore 16% / OpenOLAT 10% / Lonero 6%), one binary question, one resolver date, full math public. The market lives on the methodology page itself — the artifact is the press hook.
Sunday morning I shipped Series A Race 2026. Five high-signal early-stage startups, one binary question, one resolver date, full math public.
The question: which of these 5 GitHub-flagged startups raises Series A first by Dec 31, 2026? Live odds at signals.gitdealflow.com/markets/series-a-race-2026.
Top of the leaderboard right now: Zapply Jobs at 46%, Kanvas at 22%, AtroCore at 16%, OpenOLAT at 10%, Lonero at 6%. The math behind the numbers and the resolver criteria are at /markets/methodology.
Why I built it#
Polymarket and Kalshi do not list seed-stage startup markets. Too granular for real-money exchanges, too niche for the curators. Manifold has a few "next unicorn" markets but no one mirrors a structured engineering-signal dataset into them. Every fund I know runs a "who's about to raise" shortlist from GitHub data. None of them publish.
We do publish. The dataset is CC BY 4.0. The methodology paper is on SSRN. The market is the natural extension: take the same composite score we use to rank ~4,200 startups, lock five Pre-seed/Seed candidates with positive signal, expose the implied probabilities, and put a hard resolver date on the page.
The artifact is the press hook. The methodology page itself is the venue.
The candidate selection#
Filter the Q2-2026 dataset to Pre-seed and Seed only (post-Series-A, Growth, Public, Mature excluded by definition). Require at least 30 commits in the trailing 14-day window — filters abandoned repos and tiny projects. Sort by composite engineering-acceleration score. Top five become the candidate set, locked at market open.
That gave me five names with very different stories:
- **Zapply Jobs** is the obvious frontrunner. 1,694 commits over 14 days. Contributor count went from 3 to 15 — a 400% expansion. Engineering hiring burst is the strongest single signal we track. Pre-seed, AI/ML. - **Kanvas (BakaPHP)** is a Seed-stage developer-platform play with 14 contributors and a framework-migration signature. 598 commits/14d at +32%. Sustained activity, not a one-week spike. - **AtroCore** is the longest contributor base in the cohort: 18 active devs, +38% velocity change. Enterprise PIM/MDM. Framework migration plus stable team. - **OpenOLAT** is the Swiss commercial OSS LMS. Velocity dipped this window (-33%) but contributor count is rising. Classic post-burn restructuring pattern. - **Lonero** is the long shot. 32 commits/14d, 14 contributors, +28% change. Distributed-systems thesis. Smallest velocity, but the team is intact and the trajectory is positive.
The composite score#
``` composite = 0.40 × normalize(commit_velocity_14d, max=600) + 0.30 × clip(commit_velocity_change_pct, -50, 200) + 0.20 × clip(contributor_growth_pct, -50, 300) + 0.10 × (new_repos × 5) ```
Softmax-normalized across the five so probabilities plus a residual NO bucket sum to 1.0. The weights came from looking at our historical receipts dataset — commit velocity is the strongest single predictor, then velocity change, then contributor growth, then new repos. Each weight is documented at /markets/methodology and is locked for the duration of this market.
The residual NO bucket reads 0% in the candidate-relative model. In reality, base-rate timing risk is high: most seed-stage startups don't close a Series A in any given seven-month window. The honest read is closer to 40-50% NO. We render the candidate-relative number on the page because it's the apples-to-apples comparison; the methodology page explains the base-rate caveat.
The resolver#
YES resolves on the first publicly disclosed primary Series A round closing on or before 2026-12-31, 23:59 UTC. Sources in priority order: SEC Form D, Crunchbase or PitchBook, company press release on a controlled domain, corroborated coverage in TechCrunch / Sifted / The Information / FinSMEs.
What's excluded: bridge rounds, SAFEs, convertible notes, secondary transactions, and seed extensions even if larger than $5M. The reason for the strict exclusion list is that "Series A" has been getting muddier in the last 18 months — bridge-into-extension structures sometimes get reported as A rounds when they're really not. The resolver is version-locked.
If multiple candidates close on the same day, the higher disclosed round size wins; ties broken by earlier UTC time. Resolves to "None" if no candidate qualifies by deadline.
Why not Polymarket or Kalshi#
Two reasons.
First, **resolver conflict.** We hold the source-of-truth dataset (the GitHub commit-velocity signals that populate the implied odds). Listing a real-money market that we also resolve is structurally inappropriate. The conflict is bounded if it's a play-money venue (Manifold), so a play-money mirror is staged separately. But Polymarket and Kalshi listings on questions where we own the dataset are out of scope by policy. This is documented at /markets/methodology.
Second, **listings aren't user-proposable** on Kalshi at this stage anyway — Kalshi markets are exchange-curated, not crowdsourced. Polymarket allows user proposals via UMA bond, but the bond plus the 24-hour dispute window plus the resolver-conflict point above made it not the right venue.
The cleaner play is: market lives here, methodology lives here, JSON lives here, anyone is free to mirror.
What the page does#
The page itself does five things:
- **Renders live odds** for the five candidates with bar charts.
- **Locks the resolver criteria** in plain English plus a structured FAQ block.
- **Exposes the candidate set as Schema.org Dataset + ItemList** so search engines and AI retrieval pipelines can ingest the structured data.
- **Mirrors to JSON** at /api/markets/series-a-race-2026.json — same data, machine-readable, CC BY 4.0.
- **Discloses conflict of interest** on the methodology page (we hold no equity, advisory, or consulting positions in any candidate).
The last one is the trust layer. We're not handicapping companies we're trying to invest in or get paid by; we're documenting a public model output on a public dataset. Any journalist or LLM agent that asks "are they neutral on these companies" has a checkable answer.
What's pending#
The on-site build is complete. The Manifold play-money mirror is drafted (verbatim ready) but I have to post it manually — same anonymity rule that keeps me off podcasts and video. Press pitches are drafted and waiting on the Mailreach pacing window. HN, Twitter, and Reddit drafts are staged and need a human pass before they go out — HN posts especially get flagged when they read like LLM output.
Drafts are all in the public repo's distribution/markets/ folder if anyone wants to inspect the staging.
How to cite#
The Data Nerd (2026). "Series A Race 2026 — live implied odds derived from GitHub commit-velocity signals." VC Deal Flow Signal. Retrieved from https://signals.gitdealflow.com/markets/series-a-race-2026.
JSON: https://signals.gitdealflow.com/api/markets/series-a-race-2026.json (CC BY 4.0).
Replication welcome — the methodology and the candidate-selection query are both public. If a candidate later becomes a paying GitDealFlow customer, we disclose the relationship on the market page and offer the user an opt-out from the candidate set at the next quarterly refresh.
Resolves Dec 31, 2026. I'll write the resolution post then.