The Omni-Compass LLC · Live Benchmark Engine
The Omni-Compass Benchmark Engine
Real-time four-arm benchmark · SHA-256 sealed · Monte Carlo validated
Run Live Benchmark EngineLive Validation
Benchmark Engine — omni-v26
fingerprint
$ omni-compass --engine=omni-v26 --fingerprint= --mode=benchmark --validate --arms=4
Loading admissibility constraints from realms/catalog_v5.csv...
Verifying Python ↔ C++ parity (17 modules, SHA-256)...
✓ Byte-for-byte parity confirmed — all 17 twin pairs
Initialising 4-arm benchmark on GitHub Actions runners...
· Arm A — the chip native — baseline
· Arm B — the chip with Omni-Compass on native — the runs disagree
✓ Omni index GAIN +20.0% (work, speed and machines · GitHub's machines)
✓ Arm D — one run — 99th-pct GAIN −11.0% beyond the noise
Running Monte Carlo simulation (10,000,000 iterations)...
✓ Basin geometry valid — closed-circle invariance confirmed
✓ Supervisory governor active — admissibility envelope enforced across all 10 domains
→ ENGINE ADMISSIBLE. Omni-Compass omni-v26 [] — all stated admissibility criteria satisfied under V26 conditions.
Four-Arm Results — omni-v26
| Arm | Name | Result | Place | Verdict |
|---|---|---|---|---|
| A | EVEN | |||
| B | EVEN | |||
| C | EVEN | |||
| D | EVEN |
Source: ·
ARM B · THE CHIP WITH OMNI-COMPASS ON NATIVE
the same card, the same rented window
the runs disagree
Runs: +1.46% / +0.73% / -0.69%
Arm B, the chip with Omni-Compass on native, against arm A, the chip native (one A10, three runs, omni-v23): the work for the bill up 1.5%, up 0.7%, down 0.7%: the runs disagree, not yet shown to rise.
omni-v23 · results/TWO_BILLS.md (results/gpu/V23_A10_CAP_FULL_ABC.md)
Full 7-tab benchmark report
Methodology · Per-Category · GPU · Receipts & SHA-256 · Run IDs
LIVE ENGINE · omni-v26
Activate the Engine.
This is the actual Omni-Compass governing engine — not a demo, not a simulation. Press Run to execute the 4-arm benchmark validation live. Every result is reproducible from the public GitHub repository.
Live Benchmark Engine
Not a demo. Not a simulation. The actual Omni-Compass 4-arm benchmark — reproducible from the public GitHub repository.
ARM A
Native baseline — reference arm
Work GAIN
—
Latency Δ (−=faster)
—
Unmodified native environment — no Omni governance. All other arms are measured against this baseline.
ARM B
Omni direct — resource GAIN vs baseline
Work GAIN
—
Latency Δ (−=faster)
—
Omni governance applied directly. Measures resource GAIN over the native baseline under standard operating conditions.
ARM C
Whole-system ON-SITE — full governance
Work GAIN
—
Latency Δ (−=faster)
—
Full Four-Stack Unified Law under maximum load. Whole-system GAIN on-site with bare-metal Kubernetes governance.
ARM D
the cloud with Omni-Compass on native
Work GAIN
—
Latency Δ (−=faster)
—
V26 extended arm. GPU bare-metal configuration — governing law applied to H100 hardware. Arm D provides the strongest available isolation by running without measurement instrumentation overhead.
Monte Carlo Simulation
— / 10,000,000+20.0%
Omni Index GAIN
Work, speed and machines · GitHub machines
+22.2%
Kubernetes
Scheduler · CPU layer · 9 scenarios
−1.3%
GPU / Chip
A10 card · own meter · GAIN all 3 runs · omni-v23
the runs disagree
Arm B Result
Chip + Omni-Compass · three runs
10M+
Monte Carlo Runs
99.9% CI · sealed
CACHED
GitHub Sync
omni-v26 · fallback values · check GitHub
1,107
Muscles/Nodes Governed
full catalogue — GitHub verified
Every result shown here is independently reproducible. Clone the repository, run the validation harness with the published seed, and compare results.
LIVE ENGINE
The Machine Is Running
Every gear, every pulse, every orbit — governed in real time. This is what a live engine looks like.
Every commit
Build Frequency
Auto-built from source
+20.0%
Whole-System GAIN
Current benchmark
—
Core SHA-256
Nov 12, 2026
—
Live Test Count
From OMNI_INDEX.json
The Live Engine Manual is the engine's own documentation — generated directly from source on every commit to the GitHub repository. It is not written by hand and not manually updated. Every section reflects the current state of the running engine: the current version tag, the current SHA-256 seal, the current benchmark results, and the current module inventory. It is the most authoritative technical reference for the engine as it runs right now.
Auto-Built on Every Commit
The manual is regenerated on every push to the main branch. The build pipeline extracts documentation from source docstrings, benchmark results from OMNI_INDEX.json and ABC_TABLE.json, and seal data from SEAL.json. No manual editing, no stale sections.
Version-Tagged on Every Push
Every build of the manual is tagged with the version string, the core SHA-256, and the seal timestamp. The tag is embedded in the manual header and in the SEAL.json manifest.
Never Manually Edited
The Live Engine Manual contains no manually written sections. Every sentence is generated from source: docstrings, type annotations, configuration schemas, and benchmark output. If the source changes, the manual changes. If the source is sealed, the manual reflects the seal.
Current Benchmark Numbers
The benchmark sections of the manual are populated directly from OMNI_INDEX.json and ABC_TABLE.json on every build. The whole-system GAIN and Kubernetes throughput gain shown in the manual are the same numbers shown on this site — both sourced from the same JSON files.
Module Inventory
The module inventory section lists every Python module, every C++ component, and every seal twin pair in the current deployment. The inventory is extracted from SEAL.json on every build.
Commit-Level Traceability
Every section of the manual is traceable to the commit that produced it. The manual header includes the commit SHA, the build timestamp, and the seal manifest hash. Auditors can verify that the manual matches the deployed implementation by checking the seal.
EXTERNAL REFERENCES