Benchmark Engine Results — Four-Arm V25 Verified GAIN
Four-arm controlled benchmark: +12.9% whole-system GAIN bare-metal, +5.0% through-cloud, +28.0% Kubernetes scheduler. 10M Monte Carlo, 99.9% CI. SHA-256 sealed.
HOW TO READ EVERY NUMBER ON THIS PAGE
A Plus Is Not Always a Gain. A Minus Is Not Always a Loss.
Every metric on this page is judged by one question: does this help the system or hurt it? The answer depends on what the metric measures — not just whether the number went up or down.
THROUGHPUT & WORK COMPLETED — HIGHER IS BETTER
A plus (+) means Omni completed more work per dollar billed. Same invoice. More output. That is a real gain — the system is thriving. The master equation: Work Completed ÷ Bill = Work per Dollar. The bill does not drop. The work goes up.
LATENCY, ENERGY, VIOLATIONS — LOWER IS BETTER
A minus (−) means Omni reduced latency, energy draw, or violation count. Less of these is better — the system does the same work for less cost or time. A minus here is a win. The verdict badge reads FASTER, not GAIN, so you always know which direction is the improvement.
EVEN — OMNI ROUTED AWAY
An EVEN verdict means Omni-Compass detected it could not improve this path and stepped aside. The system held at parity — no degradation, no loss. A plus or minus on an EVEN metric means the baseline handled it; Omni did not intervene. This is by design: the engine never governs a path it cannot improve.
GITHUB VERIFICATION — EVERY NUMBER IS LIVE
Every number on this page is fetched live from the public GitHub repository on every page load. The benchmark is SHA-256 sealed, version-stamped V25, fingerprint 66c8f2d94b3e4725. All US patent, copyright, and trademark filings are on record. Owned by Omni-Compass LLC / Dubra Software.
SIX DISTINCT MEASUREMENT SCENARIOS
+12.9%
Service-Layer GAIN
OMNI_INDEX weighted · EVENs included · measured at system boundary
All 5 operations active simultaneously
No cloud overhead · cluster boundary only
SHA-256 fingerprinted · reproducible
+5.0%
Kubernetes → Cloud → Bare Metal GAIN
Full network path · cloud overhead included · ingress + egress
Same governance stack · cloud overhead included
Arm C · full-stack composition
99.9% confidence interval
+28.0%
Kubernetes Scheduler GAIN
Orchestration layer isolated · 9 scenarios · avg across all runs
Drop-in governor · zero app changes
GitHub Actions · real cluster runs
Range varies across scenarios — see Per-Category tab for live data
Zero Drift
Closed-Circle State Evolution
Zero inadmissible state transitions · convergence basin never breached
Convergence basin maintained throughout
Four-Stack Unified Law governs all allocation
Arm B · drop-in operator · zero app changes
9
Domains Governed
Aerospace · Defense · Energy · Enterprise · and 5 more
Same governing convergence control plane law across all domains
24 parity modules verified
Scale-invariant · no domain-specific tuning
10M+
Monte Carlo Iterations
99.9% confidence interval · every run
SHA-256 fingerprinted on every result
Reproducible · sealed · open on GitHub
Four-arm controlled · independently verified
DEPLOYMENT CONTEXT COMPARISON
Same governing convergence control plane law · two deployment contexts · independently measured. GAIN is not additive across arms — each number is a direct measured outcome vs the Arm A baseline.
ON-SITE · BARE-METAL
No cloud overhead · cluster boundary only · abc.web
Arm B
+5.0%
Direct governor
Arm C
+12.9%
Full Four-Stack
Arm C adds full Four-Stack composition on top of Arm B governance · both measured vs Arm A baseline
THROUGH-CLOUD · FULL NETWORK PATH
Ingress + egress + inter-service + external API · abc.multi
Arm A
baseline
No governance
Arm C
+5.0%
Full Four-Stack
Same governing convergence control plane law as on-site · cloud overhead reduces measured GAIN · both are real, independently verified outcomes
Complete Results Table
| Metric | Arm A | Arm B | Arm C | Arm D | Verdict | Note |
|---|---|---|---|---|---|---|
| Whole-System Work GAIN | baseline | — | +12.9% | — | GAIN | Primary outcome metric |
| Service-Layer GAIN | — | — | +12.9% | — | GAIN | OMNI_INDEX weighted · EVENs included |
| 95th-pct Latency Δ | 124.8 ms p95 | — | −11.8% p95 | — | GAIN | Lower is better |
| Admissibility Rate | 100% | 100% | 100% | — | GAIN | Zero inadmissible transitions |
| Convergence Rate | 100% | 100% | 100% | — | GAIN | All runs converged |
| Parity (Python↔C++) | ✓ | ✓ | ✓ | ✓ | GAIN | Byte-for-byte SHA-256 |
| Monte Carlo Validity | ✓ | ✓ | ✓ | ✓ | GAIN | 10M iterations, 99.9% CI |
| Closed-Circle Invariance | ✓ | ✓ | ✓ | ✓ | GAIN | Basin never violated |
Reproduce these results
Clone the repository. Run the validation harness with seed 0x7F3A9C2E. Compare results line by line. Every number on this page is reproducible.