Empirical Verification
Telemetry & Benchmarks
Measured gate latencies, token economics, and reproducible offline test suite verification.
The 22/22 Offline Integrity Suite
Guarantees that skills, frontmatter headers, hook bindings, and confidence policies remain intact without external cloud dependencies.
$ bash scripts/test.sh
Running jev-superpowers offline verification suite...
✔ jev-using-superpowers: valid frontmatter
✔ jev-brainstorming: valid frontmatter
✔ jev-writing-plans: valid frontmatter
✔ jev-executing-plans: valid frontmatter
✔ jev-systematic-debugging: valid frontmatter
✔ jev-verification: valid frontmatter
✔ Cross-platform installers verified (install.sh, install.ps1)
✔ hooks.json wires Stop gate
✔ hooks.json wires PreToolUse gate
✔ pre-commit blocks on git-jev FAIL
✔ stop gate executable
✔ pre-commit ignores plain ls
✔ session-start injects Jev router
✔ installer fails without key
✔ jev-using-superpowers documents failure modes
✔ jev-brainstorming documents failure modes
✔ jev-writing-plans documents failure modes
✔ jev-executing-plans documents failure modes
✔ jev-systematic-debugging documents failure modes
✔ jev-verification documents failure modes
✔ confidence policy exists
✔ README avoids 0.0% absolute
Test results: 22 passed, 0 failed. Total execution time: 1.4sLatency Percentiles
Gate Reflex Latency vs Generative LLM Judges
Testing 500 consecutive code verification requests on an Intel Core i3 environment.
| Gate / Operation | p50 Latency | p90 Latency | p99 Latency | Payload Size |
|---|---|---|---|---|
jev-guard (Shell check) |
68ms | 82ms | 104ms | 0.4 KB |
jev-scout (Crate scout) |
82ms | 98ms | 122ms | 1.2 KB |
git-jev (Diff verification) |
94ms | 118ms | 148ms | 4.8 KB |
limpet (Stop verification) |
76ms | 92ms | 116ms | 2.1 KB |
laya (Local FOSS Engine, 421M) |
33ms | 38ms | 52ms | 0.0 KB (Local) |
| LLM Judge (Generative Prompt) | 2,400ms | 3,800ms | 6,200ms | 18.0 KB |
Cost Efficiency
Token Economics: $0.042 / Million Tokens
Generative reasoning loops cost $3.00 to $15.00 per million tokens. System One typed decisions cost pennies per day of continuous coding.
Because Jev decisions emit zero conversational tokens, a full day of 400 gated agent actions consumes less than 80,000 tokens total, costing approximately $0.0034 per developer-day.