23 times in packages/next, a PR sat without a first reviewer while the author moved on to the next task. In app-render, 20 changes landed in shared files. The authors never looked at each other's diffs.

When review bottlenecks compound with coordination gaps, the effect is multiplicative. An unreviewed PR in a directory where others are actively committing is not just slow. It is a collision waiting to happen. Cortex's 2026 benchmark found teams in this pattern see 23.5% more incidents per PR.

packages/next accumulated 23 review bottleneck windows. PRs waited without a first reviewer while the codebase kept moving.

23 detection windows flagged G1_ReviewBottleneck on packages/next. The pattern is consistent: a PR opens, no reviewer is assigned, the author context-switches, and the PR ages while new commits land around it.

app-render had 20 coordination gaps. Multiple engineers editing shared rendering files with no cross-review.

20 detection windows flagged G4_CoordinationGap on packages/next/src/server/app-render. The server-side rendering path is high-stakes code. Changes from different authors landed without mutual review.

21 velocity spikes detected. Bursts of merge activity followed by slowdowns, with no visible team response to either.

21 detection windows flagged G3_VelocityChange with velocity_spike pattern. The merge rate oscillated between high-throughput bursts and near-stalls.

acdlite had 3 review bottleneck windows. A core maintainer's PRs waiting for review signals a process gap, not a people gap.

3 detection windows flagged G1_ReviewBottleneck on acdlite as author. When a principal engineer's PRs wait for review, the team has a structural problem.

The app-render coordination gap was detectable on the first day two authors had open PRs in the same directory. Caught then, it takes a 5-minute Slack message to resolve. Caught after merge, it takes a sprint.

Analysis of 82 situation snapshots from vercel/next.js. ARC detection types: G1_ReviewBottleneck, G3_VelocityChange, G4_CoordinationGap. Detection window: 14-day rolling lookback.

The patterns above were detectable within 24 hours of the first signal. Your AI agent opened files in the same area. It did not know.

Zamski's MCP server gives every agent on the team coordination context before it writes code. One setup.

Want to see this for your repos?