Coordination findings from public repositories

ARC analysis of 25,991 pull requests across 5 public repositories. The same 3 coordination failures repeated. Review bottlenecks. Velocity shifts. Engineers editing the same files without talking to each other. Every PR number links to GitHub.

Across 25,991 PRs analyzed, 12.1% waited more than 48 hours for first review. The p90 review time was 57.39 hours.

Caught on day 1, a review bottleneck takes a 30-second reassignment. Caught on day 5, the author has context-switched and the PR needs a full re-read.

vercel/next.js:packages/next had reviews piling up without a first reviewer across 23 detection windows. See the next.js report

123 coordination gaps detected across 5 repos. In facebook/react:packages/shared, hoxyq, eps1lon, and jackpope all edited the same files. None reviewed each other's PRs.

Caught on day 1, a coordination gap takes an hour to address. Caught on day 14, it takes a sprint.

G4_CoordinationGap on packages/shared with 3 PRs in the same 14-day window: See the React report

85 velocity anomalies detected. facebook/react's merge rate shifted 60 times in the analysis window. vercel/next.js spiked 21 times.

A velocity shift detected the same day gives the team lead time to rebalance load. Detected after a sprint, it shows up as a missed deadline.

The velocity changed. The team did not notice. The pipeline kept moving. Each shift is a missed intervention point.

Across 5 public repos, the same coordination failures repeated. Most have no human-visible symptom until something breaks.

Visible patterns are manageable. Invisible ones compound until they surface as incidents.

The grounding absence:

These coordination gaps repeated across every repo we analyzed. Your AI agent opens a file, writes code, and creates a PR. It does not know that two other engineers edited the same module this week.

Zamski's MCP server changes that. One setup. Every agent on the team gets coordination context before writing code.

Want to see this for your repos?