Synthetic workspace with 1110 projects arranged in a 3-level fan-out (10 groups x 10 subs x 10 leaves). Each project defines build, copy, and cat targets that operate on a shared lorem.md file — no real compilation, just enough I/O to exercise Nx's task pipeline.

Prerequisites: hyperfine (cargo install hyperfine)

# Run all benchmarks and compare against goals + baseline
pnpm nx run benchmarks

# Run a single benchmark
pnpm nx bench:version benchmarks
pnpm nx bench:show-projects benchmarks
pnpm nx bench:cat-warm benchmarks
pnpm nx bench:copy-warm benchmarks

Each bench:* target depends on ^build so the Nx packages are compiled first.

All benchmarks use NX_NO_CLOUD=true, run nx reset before each iteration, and collect at least 5 runs (10 for version) via hyperfine.

Performance is tracked with two files:

• goals.json (committed) — target times the team agrees on. CI fails if a benchmark exceeds its goal.
• baseline.json (gitignored) — your local machine's numbers for personal comparison.

The run-benchmarks.ts script reads both and prints a table with colored deltas showing how the current run compares to each.

# First run auto-creates baseline.json
pnpm nx run benchmarks

# Explicitly overwrite with fresh numbers
pnpm nx run benchmarks -- --set-baseline

1. Each bench:* script invokes hyperfine and writes a results-<name>.json file.
2. The run target depends on all bench:* targets, so they execute in sequence (parallelism: false).
3. After all benchmarks finish, run-benchmarks.ts reads the result files and prints the comparison table.

Benchmarks run as part of the affected target pipeline in CI (nx affected --targets=...bench). The goals in goals.json act as the regression gate.