KLOUDY / quantum
Quantum security research, in the open.
Track Kloudy’s quantum-security experiments through dated aggregate results. Compare QPU, hybrid and OS catch rates, dormant-actor detection and predictability results. The table includes both trajectories, classical controls, sample sizes and paired p-values. It is a saved research record, not live protection or an established quantum-advantage claim.
Research record · Aggregate findings only. No QPU connection is needed to read these pages.
Building with an IDE? Use the CLI and connect one Kloudy MCP ↗
QPU vs OS catches in the final scripted trajectory; 121 actors. Paired p = 0.0759.
The QPU arm caught the dormant actor at step 18; the OS arm at step 29. One dormant actor, one trajectory.
Caught by QPU and missed by OS in that trajectory. OS also caught 20 actors that QPU missed.
Hardware-backed research in a scripted simulation. These are measured observations, not a claim of deployed protection or established quantum superiority. The final QPU/OS paired p-value is 0.0759. At the 5% threshold, a significant difference is not established. The equal-budget control caught 92; the hybrid caught 85.
Dated catch results
n is the number of actors in a scripted world, not independent experiments. There are two trajectories: all pre-annotation comparisons share the first; all final comparisons share the second. Six comparison rows do not mean six independent runs. Dates are the report completion date. “Higher catch” means the quantum-containing arm caught more; it does not mean a proven quantum advantage.
| Reported / comparison | Runs / n / attacker set | Classical catch rate | Quantum-containing catch rate | Paired p-value | Finding |
|---|---|---|---|---|---|
| 2026-09-29 Pre-annotation | 1 / 120 One scripted world; 120 moving actors | OS: 51.7% (62/120) | QPU: 51.7% (62/120) | 1 | Higher catch: No. Significant difference: No. Matched inspection budget. Quantum advantage: Not established. |
| 2026-09-29 Pre-annotation · unequal budget | 1 / 120 One scripted world; 120 moving actors | OS: 51.7% (62/120) | OS + QPU: 85.8% (103/120) | Not reported | Higher catch: Yes. Significant difference: Not reported. Twice the classical arm’s inspection budget. Quantum advantage: Not established. |
| 2026-09-29 Pre-annotation · matched budget | 1 / 120 One scripted world; 120 moving actors | Dual OS: 82.5% (99/120) | OS + QPU: 85.8% (103/120) | Not reported | Higher catch: Yes. Significant difference: Not reported. Matched inspection budget. Quantum advantage: Not established. |
| 2026-09-29 Final annotated | 1 / 121 One scripted world; 120 moving actors and one dormant actor | OS: 41.3% (50/121) | QPU: 52.9% (64/121) | 0.0759 | Higher catch: Yes. Significant difference: No. Matched inspection budget. Quantum advantage: Not established. |
| 2026-09-29 Final annotated · unequal budget | 1 / 121 One scripted world; 120 moving actors and one dormant actor | OS: 41.3% (50/121) | OS + QPU: 70.2% (85/121) | 0.0000021 | Higher catch: Yes. Significant difference: Yes. Twice the classical arm’s inspection budget. Quantum advantage: Not established. |
| 2026-09-29 Final annotated · matched budget | 1 / 121 One scripted world; 120 moving actors and one dormant actor | Dual OS: 76.0% (92/121) | OS + QPU: 70.2% (85/121) | 0.3604 | Higher catch: No. Significant difference: No. Matched inspection budget. Quantum advantage: Not established. |
Unpredictability experiment
Reported 2026-09-29. Each source was evaluated on 2048 held-out selections. These are attacker prediction accuracies, not catch rates; lower is better. QPU, hybrid and OS sources all passed the study’s predictability criteria. This experiment did not establish an advantage over the OS source.
| Source | Frequency attacker | First-order attacker |
|---|---|---|
| Fixed | 6.25% | 100% |
| Weak PRNG | 6.396% | 6.787% |
| Windows CSPRNG | 6.494% | 5.859% |
| QPU-direct | 6.982% | 6.641% |
| Hybrid | 6.152% | 5.615% |
The fixed and deliberately compromised weak-generator controls were predictable under their informed attacks. That does not describe a properly used OS cryptographic source. A stolen precomputed schedule is readable regardless of its origin.
What the lab still needs to establish
Independent attacker populations, repeat runs and production traffic are still needed. The main simulated scanner’s flagged-clean rate was too high for operational alerts. No quantum advantage or live network enforcement is established by this snapshot. The QPU stays off the browser’s production request path.
Follow the evidence
Download the imported aggregate record (JSON) · Full QVault-Test research record (repository access required) ↗
The download contains all four arms from both catch trajectories and the five main prediction variants from the two imported experiment reports, report fingerprints and the source revision. Other lab experiments are outside this snapshot. No raw measurements, credentials or implementation details are published.
Results last updated: . This is a saved research snapshot, refreshed during releases. It remains readable if the lab or refresh service is unavailable; it is not a live QPU status monitor.
Less setup. More useful work.
For IDEs and bots: one CLI, one MCP, tools appear when needed.
Find with SEO/GEO, read FHIR for medical data or KLDY for other information, then select one scoped Toolbox tool when action is needed.
Passive local reads use no model calls. Narrowing 100 candidate definitions to 10 would mean 90% fewer definition fetches in that example; this is an illustration, not measured token savings.