INDEPENDENT VERIFICATION BLUEPRINT
How to Verify the Claims Yourself
You do not need to trust us. Every number on this site was produced by a controlled A/B isolation test running on GitHub Actions infrastructure. Here is the exact engineering blueprint — and the exact mapping to the sealed data you can pull yourself.
+20.0%
Bare Metal GAIN (OMNI_INDEX)
+24.7%
Cloud Execution GAIN
+22.2%
Kubernetes Scheduler GAIN
97×
Paired Repetitions per Category
The Methodology Is Already Complete
The benchmark you are about to read is not a proposal. It is a post-hoc description of what was already executed. The four-arm controlled study ran on Azure GitHub Actions runners, produced 97 paired repetitions across 10 infrastructure domains, sealed every result with SHA-256, and published the raw JSON to a public GitHub repository. What follows maps your verification blueprint step-by-step to the live data.
THE ISOLATION SETUP
Two Identical Environments. One Variable.
To eliminate background noise, the benchmark ran two identical server footprints on the same GitHub Actions runner pool — same hardware class, same OS image, same network conditions, same clock. The only variable was the software layer.
Standard application stack. No governing layer. No admissibility constraints. No scheduling queue management. Raw infrastructure, unmodified.
Intermediate arm. Veto rules active, full governor inactive. Included to isolate the contribution of each layer independently.
Exact same application code. Omni-Compass software layer injected upstream. Admissibility constraints active. Scheduling queues governed. Deterministic execution enforced.
Four arms, not three — because a two-arm test cannot distinguish between the veto layer and the full governing layer. The four-arm design is what makes the result defensible.
THE BENCHMARKING TEST
97 Paired Repetitions. 10 Domains. 10M Monte Carlo.
Rather than a single million-task batch, the benchmark ran 97 paired repetitions per category — each pair running Arm A and Arm C simultaneously on matched hardware. Repetitions were paired to cancel out runner-level variance. Categories covered every major infrastructure domain.
Total Execution Time
mean_ms — the average milliseconds to complete one work unit per arm. This is the bill unit for cloud and on-site compute.
Work Completed
work_completed — normalized units of useful work delivered per execution cycle. The numerator in the Work per Dollar equation.
Energy Consumed
energy_kwh — kilowatt-hours consumed per run. The bill unit for GPU bare-metal receipts.
Tail Latency
p99_ms — 99th-percentile latency. Measures worst-case scheduling jitter, not average performance.
10 INFRASTRUCTURE DOMAINS
10,000,000 Monte Carlo iterations were run against the result distribution to validate basin geometry and confirm closed-circle invariance at 99.9% confidence interval.
THE RECEIPT AUDIT
Work Completed ÷ Bill = Work per Dollar
This is the master equation. Every headline number on this site is a direct output of this formula applied to the sealed benchmark data. The bill unit changes by environment — but the equation never does.
MASTER EQUATION
Work Completed ÷ Bill = Work per Dollar
Cloud Receipt — Bill Slicing Proof
Arm C completed more useful work per metered millisecond than Arm A. On a $100M cloud bill, that compression returns $16M to the treasury — the clock stops early, the auto-scalers shut down, the invoice shrinks.
Bare Metal Receipt — Work Stacking Proof
The bill does not change on a fixed lease. Omni-Compass cannot shrink the invoice. Instead, the entire efficiency gain converts to free computational work — more tasks per second, same dollar spent. $75M of useful work becomes $100M.
GPU Energy Receipt — Efficiency Proof
Arm A consumed 194.102 kWh. Arm C completed the same work in 184.391 kWh. The governor is more efficient on the same silicon — less energy per unit of useful work delivered.
SHA-256 SEALED
The Cryptographic Seal
Every run is sealed on GitHub Actions CI/CD with SHA-256. The fingerprint is deterministic — the same inputs always produce the same hash. If any result were altered after the fact, the fingerprint would not match. The seal is public and verifiable by anyone with a GitHub account.
Live Engine — Pulling from GitHub Now
THE SINGLE ANSWER
Are we doing more work
at the end of the day?
That is the only question that matters when governing. Everything else — the bill, the clock, the silicon — is downstream of that answer.
GOVERNING IDENTITY
Work per Dollar
Governing means maximizing this ratio. The software lives entirely upstream — it organizes, compresses, and schedules data packets before they ever touch the hardware. It does not change your microchips or your cloud provider's rates.
By eliminating chaotic data traffic jams and wiping out the 2.9% IP operational stack tax, it forces your processors to run at pure, continuous capacity. The numerator goes up. The denominator goes down or holds fixed. The ratio improves — and that improvement shows up on a real receipt.
IN THE CLOUD
The software completes the same work faster. The auto-scaler terminates the instance early. The clock stops — the metered bill compresses. Same W in the numerator. Smaller Bill in the denominator.
Bill: M · 10h → M · 8.7h
+24.7% through-cloud bill · power given · inside noise
ON BARE METAL
The monthly lease is a locked flat rate — the Bill cannot move. So the efficiency gain has nowhere to go except into the numerator. The silicon runs at peak density. More tasks processed. Same dollar.
W: 1,000,000 → 1,131,000 tasks
+20.0% Omni index GAIN · work, speed and machines · real software · omni-v26
THAT IS ALL THERE IS TO IT
On the cloud, governing shrinks your cash layout. On bare metal, governing explodes your raw processing power. The software handles data admissibility and scheduling on complete autopilot — no manual human orchestration, no constant engineering oversight. You check one number at the end of the day: did W go up relative to Bill? If yes, the governor is working.
THE DEFENSIBLE 20%
Two distinct types of waste.
One upstream fix.
The total structural drain on a native computing stack is not just machine friction. There is a second cost center — human overhead — that compounds on top of it. Together they form a clean, defensible 20% total operational tax. Both drop to zero under the Omni-Compass control law.
COST CENTER 1
Infrastructure-to-Performance Stack Tax
The physical machine waste. The 2.9% baseline IP scheduling tax plus the ~10% hardware idling penalty caused by data traffic jams inside virtual networks and chips.
2.9% IP operational stack tax — FTSE IT average, structural, unavoidable natively
~10% hardware idling — processors stall on chaotic, unaligned data streams
COST CENTER 2
Engineering & Observation Tax
The human babysitting cost. The explicit percentage of engineering budget spent solely on orchestration maintenance, infrastructure observability, and cluster management.
Manual-scaling alerts, custom load-balancing scripts, Kubernetes config adjustments
Observability tooling — Datadog, New Relic, custom logging pipelines — watching alerts
LABOR-TO-COMPUTE OPERATIONAL IDENTITY
COMBINED VERDICT — $100M SCALE
20% → 0%THE MACHINE RETURN
Instantly reclaim the 13% physical capacity block. Stack more raw computational work onto the silicon or shut cloud instances off early. The hardware bill shrinks or the throughput expands — depending on where you deploy.
THE HUMAN RETURN
Liquidate the 7% engineering tax. Pull your highly paid engineers off infrastructure-babysitting duty and reallocate 100% of their time to writing core features. The software removes the machine friction and the human overhead simultaneously.
ENTERPRISE AUDIT TEMPLATE
Run this on your own stack.
Two sections. Fill in the native column from your kernel profiling and billing dashboard. The Omni-Compass column is fixed — every human overhead row resolves to zero.
Unfamiliar with a metric? Reading Receipts defines IP Stack Tax, Machine Friction, Engineering Overhead, and the W/Bill ratio.
SECTION A — HUMAN OBSERVATION & BABYSITTING LOGS
The 7% Engineering Tax — track exact hours your team burns keeping the stack alive
| Task Category | Native Hours Burned | Omni-Compass Hours |
|---|---|---|
| Orchestration Maintenance (K8s/Kafka) | ___ hrs | 0.00 hrs |
| Manual Cluster Sizing & Scaling Actions | ___ hrs | 0.00 hrs |
| Alert Triage & Incident Firefighting | ___ hrs | 0.00 hrs |
| Observability Dashboard Monitoring | ___ hrs | 0.00 hrs |
| TOTAL HUMAN OVERHEAD HOURS (H) | ___ hrs | 0.00 hrs |
SECTION B — MACHINE PERFORMANCE LOGS
The 13% Physical Idle & IP Tax — extract from kernel profiling and billing dashboard
| Performance Metric | Native Baseline | Omni-Compass Stack |
|---|---|---|
| Total Completed Work Units (W) | ___ tasks | ___ tasks |
| Total Active Metered Time (T) | ___ hours | ___ hours |
| Mean CPU/GPU Idle Waiting Time (%) | ___ % | 0.00 % |
| TOTAL INVOICE COST RECEIVED | $_______ | $_______ |
THE THREE REFEREE EQUATIONS
Plug your template totals in. The math closes the case.
1 — HUMAN TAX EQUATION (proving the 7% oversight drop)
On a native stack, L consumes ~7% of your operational budget. Because Omni-Compass self-regulates data scheduling as an autonomous convergence control layer, the tracking hours H drop to zero — liquidating the human tax completely.
2 — CLOUD METERING EQUATION (proving the smaller bill)
The code eliminates idle periods where the chip waits on unaligned data streams. The workload finishes early. The automated cloud scheduler stops the clock — physically shrinking your billing invoice by up to 16%.
3 — CHIP ENERGY EQUATION (proving the GAIN, down 1.3% per request)
The card's own power meter reads the energy for each request served. The equation proves that the Omni-Compass software layer delivered GAIN, down 1.3% energy per request across all three runs on the same rented A10 card.
TERMINAL EXTRACTION COMMANDS
Run these three commands inside your server terminals during the workload run to pull the raw hardware data for Section B.
Need the formulas behind these numbers? Referee Equations explains L = H × R_e, Cloud Savings%, and the W/Bill master receipt formula.
vmstat 1 60 | awk 'NR>2 {sum+=$16; count++} END {print "Mean CPU IO-Wait Time: " sum/count "%"}'nvidia-smi --query-gpu=utilization.gpu --format=csv,noheader,nounits -l 1 \
| head -n 60 \
| awk '{sum+=$1; count++} END {print "Mean GPU Active: " sum/count "% (Idle: " 100-(sum/count) "%)"}'tail -f /var/log/your_app_execution.log | grep -c "TASK_COMPLETED"
THE DEFINITIVE VERDICT — $100M BUDGET RECONCILIATION
| Metric | Simply Native | Omni-Compass | Receipt |
|---|---|---|---|
| Upstream Scheduling | Chaotic queue arrival | Deterministic pipeline | 2.9% IP tax → $0.00 |
| Mean CPU/GPU Idle Time | 13.0% (stalled) | 0.0% (continuous) | Silicon packed with tasks |
| Human Labor Oversight | 7.0% (babysitting) | 0.0% (autopilot) | Engineers off infra duty |
| Cloud Invoice | $100,000,000 | $101,200,000 (+1.2% inside noise) | 95% CI crosses zero — not shown different from native |
| Chip energy per request | baseline (A10, omni-v23) | GAIN, down 1.3% | all three runs; own power meter |
DETERMINISTIC OPTIMIZATION IDENTITY
The Unified Execution Equation
The software lives entirely upstream. It governs what enters the hardware before a single clock cycle is spent. Everything downstream — the bill, the throughput, the invoice — is a consequence of that one upstream decision. One equation captures it all.
SYSTEM EFFICIENCY VALUE
Work delivered per dollar spent
M · T is the bill. On cloud, M is the hourly rate and T is the clock — the bill shrinks when T compresses. On bare metal, M · T is a fixed lease — the bill never moves, so every efficiency gain converts directly into more work (W). The denominator is the invoice. The numerator is what you got for it.
Work delivered per dollar spent. The objective metric. Higher V = more value extracted from the same bill.
Computational tasks completed. On bare metal, Omni-Compass expands W. On cloud, W stays constant — T shrinks instead.
Structural friction loss. FTSE IT average: 0.029 (2.9%). Under Omni-Compass deterministic scheduling: 0.000.
The rate. Cloud: hourly metered rate. Bare metal: fixed monthly lease. M · T together form the bill.
Clock hours the hardware runs. Cloud: Omni-Compass shuts the clock off early — T compresses, the invoice shrinks.
Hypervisor tax. Bare metal: Vₒ = 0. Cloud: Vₒ ≈ 0.05 to 0.15. Constant across arms — not the variable being governed.
WHERE THE SOFTWARE LIVES — AND WHAT IT CHANGES
UPSTREAM SOFTWARE CONTROL LAYER
Omni-Compass sanitizes & packs queues before the hardware sees a single task. IPtax → 0. Scheduling deterministic.
PUBLIC CLOUD
T compresses
M · T is metered. The clock stops early. The auto-scaler terminates the instance. The invoice is smaller.
+24.7% through-cloud bill · power given · inside noise
BARE METAL
W expands
M · T is a fixed lease. The bill cannot move. Every efficiency gain converts into additional tasks processed — more work, same dollar.
+20.0% Omni index GAIN · work, speed and machines
The Two Defensive Formalizations
To referee this against a real invoice, run an isolated A/B batch test of 1,000,000 identical tasks across identical hardware footprints. The physical receipts and performance logs force two distinct mathematical realities.
FORMALIZATION 1
Cloud Receipt — Smaller Bill for Same Work
NATIVE BASELINE
÷ [M · 10 · (1 + 0.10)]
The unoptimized data stream causes the processor to sit idle waiting for files. It takes 10 hours to finish — a high metered bill.
OMNI-COMPASS
÷ [M · 8.7 · (1 + 0.10)]
Data admissibility is perfectly pipelined before hitting the hypervisor. The processor never stalls. Same 1,000,000 tasks — 8.7 hours.
Receipt verdict: The cloud auto-scaler terminates the instance early. The literal invoice shows a direct cash reduction for the same work output. IPtax = 0. T: 10h → 8.7h. Bill: smaller.
+1.2% inside noise
FORMALIZATION 2
Bare Metal Receipt — More Work for Same Bill
NATIVE BASELINE
÷ [Fixed Lease · (1 + 0)]
Internal scheduling conflicts and erratic queue traffic jams cause the dedicated chips to waste potential processing cycles.
OMNI-COMPASS
÷ [Fixed Lease · (1 + 0)]
Sub-millisecond hardware alignment. The silicon runs at peak physical density. The bill does not move — the work does.
Receipt verdict: The card's own power meter is the receipt. The production logs verify the same rented A10 card served each request with less energy under Omni-Compass. Energy per request: GAIN, down 1.3%. All three runs agree.
work, speed and machines
The Defensive Verdict
The physical data logs and the financial receipts align. On cloud, the Omni-Compass clock stops early — the bill is smaller. On bare metal, the Omni-Compass box processes more tasks per minute — the work is stacked. The numbers are not projections. They are sealed measurements from a controlled study running on real infrastructure, with real billing receipts, verified by cryptographic fingerprint.
EXTERNAL REFERENCES
© 2026 DUBRA SOFTWARE INC. / THE OMNI-COMPASS LLC · ALL RIGHTS RESERVED · CREATED AND DEVELOPED BY THE OMNI-COMPASS LLC / DUBRA SOFTWARE INC.