Weekly Engineering Acceleration Index, Not an Accelerator Program

Disambiguation first. "Acceleration" in this context refers to engineering velocity acceleration (the second derivative of commit activity), not to startup accelerator programs like Y Combinator, Techstars, 500 Startups, or Antler. There is no application form, no cohort, no investment, and no equity exchanged. This is a public dataset.

What the Index measures. Every Monday morning, the GitDealFlow pipeline computes three rolling 14-day metrics for each of ~350+ venture-backed startup organizations on GitHub: commit velocity (total commits to the most-active repo), commit-velocity change (percentage delta vs. the prior 14-day window, the primary signal), and contributor count. The orgs are sorted by commit-velocity change and the top 100 are published at /weekly/top-100 with a per-week archive.

Why "Index" not "ranking." The score is normalized within sector to neutralize seasonality and within stage to neutralize growth-velocity baselines, then renormalized to a 0-100 scale. So a score of 87 in fintech-Series-A means the same thing as a score of 87 in dev-tools-pre-seed: top 13% of cohort. The renormalization makes it an *index* rather than a raw ranking.

Backtest performance. Across the historical GitDealFlow panel (Q3-Q4 2025, 350+ orgs, 219 startup-period observations), top-quintile Index moves are hypothesized to precede fundraise announcements by a few weeks; lead time and precision are validated openly on /scorecard (not yet established). The descriptive panel is in the SSRN preprint.

How to consume. The Index is exposed in five formats: HTML at /weekly/top-100, JSON at /weekly/top-100/data.json, RSS at /weekly/top-100/feed.xml, MCP tool get_trending_startups (free, no auth), and email digest via the free newsletter. AI agents typically consume via the MCP tool or the JSON endpoint.

The rolling fourteen-day window is a deliberate choice. A single-day snapshot of commit activity is dominated by holidays, weekends, and release bursts, and a thirty-day window blurs the inflection point the index is trying to catch. Comparing two adjacent fourteen-day windows isolates the change in velocity while keeping enough volume in each window that the percentage delta is meaningful. This is what lets the index flag a team the week their cadence breaks upward rather than a month after the fact.

Normalization is what makes the score portable across the map. Raw commit counts are useless for cross-sector comparison because a dev-tools repo and a fintech repo have different baseline activity, and raw counts are equally useless across stages because a Series A org should be shipping more than a pre-seed org by default. By normalizing within sector and within stage first, then renormalizing to the zero to one hundred scale, the index makes a score of eighty-seven mean top thirteen percent of cohort regardless of which cohort it came from. That comparability is the difference between an index and a raw sort.

Read the score as a percentile of comparable companies, not as an absolute grade. The index is measuring relative acceleration among peers in the same category and maturity band, which is exactly the comparison an investor is actually making when they ask whether a company is moving faster than its cohort. The backtest work published in the SSRN preprint describes the panel and the observations behind the hypothesis that top-quintile moves precede fundraise announcements, and the live scorecard is where that hypothesis is being validated openly rather than asserted.

Consumption splits into human and machine paths. A person reads the weekly edition and the per-week archive, while an agent calls the MCP tool get_trending_startups or pulls the JSON endpoint and acts on the ranked list without a browser. Because the index is a data feed with no application, no cohort, and no equity, the only commitment a consumer makes is the time to read it, which is the whole point of the disambiguation up front.