Markets methodology

1. Candidate selection

Each market starts from the current quarterly dataset (e.g. q2-2026) of ~4,200 tracked startup GitHub orgs. For a Series A market, the eligible pool is filtered to:

• Stage label of Pre-seed or Seed in the enrichment dataset (post-Series-A, Growth, Public, and Mature excluded).
• Minimum base rate of 30 commits in the trailing 14-day window (to filter out abandoned repos and tiny projects).
• Sorted by composite engineering-acceleration score; top N (5 for Series A Race 2026) become the candidate set.

2. Composite signal score

composite_score =
    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)

Weights reflect the empirical signal strength observed in our historical receipts dataset (/receipts): commit velocity is the strongest single predictor, followed by velocity change, then contributor growth.

3. Implied probabilities

Composite scores across the candidate set are softmax-normalized and scaled so the candidates plus a residual NO bucket sum to 1.0. The residual NO bucket reflects base-rate timing risk: in any 7-month window, fewer than half of seed-stage startups actually close a Series A. We round to the nearest whole percent for display; raw values are in the JSON.

4. Resolver

A candidate resolves YES on the first publicly disclosed primary Series A round. Acceptable sources, in order:

1. SEC Form D (US) or equivalent public filing.
2. Crunchbase or PitchBook funding-round entry.
3. Company press release on a controlled domain.
4. Coverage in TechCrunch, Sifted, The Information, or FinSMEs (corroborated by at least one of the three above).

Excluded:bridge rounds, SAFEs, convertible notes, secondary transactions, and extensions of an existing seed round (even if > $5M).

5. Refresh cadence

The candidate set is locked at market open. Implied odds refresh with each quarterly dataset release (q1, q2, q3, q4) and on material funding news. Each refresh is logged in /changelog; superseded JSON snapshots remain accessible via /corrections.

6. Conflict-of-interest disclosure

The author and publisher do not hold equity, advisory, or consulting positions in any of the candidate startups listed in any market. We do not accept payment to add, remove, or re-weight candidates. We do not propose markets on platforms where we hold the source-of-truth dataset (Polymarket / Kalshi-style real-money markets) — only on play-money venues like Manifold where the resolver conflict is bounded.

If a candidate startup 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.

7. Not investment advice

Implied probabilities are model output from a public dataset. They are not a recommendation to fund, short, or otherwise transact in any of the listed companies. Anyone using this data for investment decisions assumes full responsibility for those decisions.