Windows Performance

Windows NTFS is the most common development environment. These benchmarks measure Hyperion’s manifest restore performance on Windows without any Git dependency — just cpSync and atomic rename.

10,000 file repo

A synthetic 10,000-file TypeScript workspace. Hyperion restores 10 dirty files per rollback cycle averaged over 3 iterations.

Metric	Value
Repo size	10,000 files
Dirty files	10
Iterations	3
Synthesis time (one-time)	31,796 ms
Avg rollback latency	62.034 ms
Samples	70.658, 74.560, 40.883 ms

Comparison to Git

On the same machine with the same 10K-file fixture:

Strategy	Platform	Avg Latency	Speedup
Git reset --hard + clean -fd	Windows NTFS	~570 ms	1.00×
Hyperion manifest restore	Windows NTFS	62.034 ms	9.19×

Git inspects the entire working tree and re-checks out every tracked file. Hyperion only copies the 10 files in the dirty set back from the read-only base.

Methodology

The benchmark script is at scripts/win-bench.mjs. It:

1. Synthesizes a base monorepo of N TypeScript files
2. For each iteration: copies the base into a working directory
3. Mutates DIRTY_COUNT tracked files with random content
4. Creates DIRTY_COUNT scratch files
5. Measures the time to restore all dirty files from the base and delete scratch files
6. Cleans up

No Git, no shell commands, no tmpfs — pure Node.js filesystem operations on NTFS. All timings use process.hrtime.bigint().

Running locally

$env:HYPERION_FILE_COUNT="10000"
$env:HYPERION_DIRTY_COUNT="10"
$env:HYPERION_ITERATIONS="5"
node scripts/win-bench.mjs

See also

• Benchmark Results — all sweep data including dirty-set scaling, repo-size independence, and agent search
• Reproduce the Benchmark — run the full benchmark suite