THE PROOF
Evidence
Every claim has a receipt. Every receipt has a hash. Every hash is public. This is the complete evidence record.
All benchmark data, sealed run receipts, SHA-256 fingerprints, and the complete benchmark harness are publicly available on GitHub for open evaluation and independent replication by any engineer, institution, or evaluator worldwide. No license is required to verify. A license is required to monetize or commercialize.
Receipted Claims
+20.0%
Whole-system gain
Work, speed, and machines. GitHub's machines. 97 paired reps. omni-v26.
OMNI_INDEX.json · headline_pct+96.5%
Work and speed gain
Same runs. Machines not scored. omni-v26.
OMNI_INDEX.json · service_headline_pct+22.2%
Kubernetes gain
9 tests. One loss: fairness −2.6%. omni-v26.
SITE_NUMBERS.json · kubernetesGain−11.0%
99th-percentile latency gain
Azure AKS. 97 paired reps. Beyond noise. omni-v18.
SITE_NUMBERS.json · alertMinutes−1.3%
GPU energy gain
A10 card. Own meter. All 3 runs. omni-v23.
SITE_NUMBERS.json · chipEnergyGainEvidence Cabinet
Sealed Runs
17 twin pairs. 97 paired reps. What was measured, how, and what the seal proves.
OpenMonte Carlo Validation
Statistical validation methodology. Confidence intervals. What the distribution shows.
OpenGitHub Receipts
Live SHA-256 fingerprints. SEAL.json. OMNI_INDEX.json. Direct links. Open to the world.
OpenIndependent Replication
Step by step. Hardware spec. Exact commands. Expected output. Any engineer can replicate.
OpenChain of Custody
Forensic provenance. Who ran it. When. What machine. What hash. What the hash proves.
OpenWhat the Evidence Does Not Prove
The benchmark receipts prove specific, bounded results under specific, documented conditions. They do not prove that every possible inefficiency is eliminated. They do not prove that results will be identical on every infrastructure configuration. They do not prove that the cloud bill will decrease — the Azure receipt shows a +1.2% bill increase inside the noise margin. The real cloud receipt is the 99th-percentile latency gain of −11.0%, which is beyond noise. Every claim on this site is bounded, labeled, and receipted. No claim is extrapolated beyond its receipt.