Secret scope & scaling
A secret scope (me.finance["_"]("key")) seals everything under it: values are stored encrypted and read through the branch key. These benchmarks measure what that costs next to the same workload on a public branch. The cost is the secret path itself: branch key derivation and cache, sealing and opening values (v3: Keccak-256 counter-mode keystream + HMAC-Keccak256 tag), stealth-boundary checks.
Measured on this.me 4.2.0 (2026-10-10, shared Linux VM, Intel Xeon, 8 vCPU, 16 GiB, Node 22.23.3), next to the pre-4.1.0 build on the same machine; the first published figures were on an Apple M2. Both runs are on the kernel benchmarks page.
Setup
Node ≥ 22.6. All commands run from Typescript/.
git clone https://github.com/neurons-me/.me.git && cd .me/Typescript
npm install
npm run build
1. Secret vs public branch (#9)
What it measures. The same workload on a public branch and on a secret branch: 600 nodes, lazy mode, write the shared factor then read one output. Median of 5 runs.
node tests/Benchmarks/benchmark.9.secret-scope-impact.test.ts
| public p95 | secret p95 | secret / public | |
|---|---|---|---|
| Apple M2, pre-4.1.0 (published) | 0.044 ms | 0.604 ms | ~14× |
| this.me 4.2.0, Xeon VM | not re-measured | 1.64 ms | (pre-4.1.0, same VM: 1.62 ms) |
Secret reads stay in the low-millisecond envelope; 4.2.0 is within 2% of the previous build.
2. Secret push vs pull (#11)
What it measures. Write cost (push) and first-read cost (pull) for public and secret branches at 100 / 300 / 600 nodes, lazy mode. Median of 5 runs.
node tests/Benchmarks/benchmark.11.secret-push-vs-pull.test.ts
Published (Apple M2, pre-4.1.0), p95 in ms:
| nodes | public write | secret write | public read | secret read | read slowdown |
|---|---|---|---|---|---|
| 100 | 0.0126 | 0.0464 | 0.0499 | 0.392 | 7.7× |
| 300 | 0.0045 | 0.0327 | 0.0275 | 0.311 | 11.1× |
| 600 | 0.0041 | 0.0355 | 0.0247 | 0.550 | 22.6× |
On this.me 4.2.0 (Xeon VM) the secret read p95 at 600 nodes is 1.34 ms (pre-4.1.0 on the same VM: 1.49 ms). Writes stay cheap in both planes; secret reads cost more and grow with node count.
3. Secret branch at scale
What it measures. Full-branch reads, cold vs warm, leaf vs branch, and snapshot rehydration, public vs secret, at 1,000 / 10,000 / 100,000 / 250,000 records with 256- and 1536-dimension embeddings (11 runs each; sizes set with BENCH_SECRET_SCALE_N and BENCH_SECRET_SCALE_EMBED).
node --expose-gc tests/Benchmarks/benchmark.secret-branch-scale.test.ts
No published results yet: this one has not been run for the docs on any version.
