FOR QUANTUM RESEARCH SOFTWARE ENGINEERS
Two patches.
Same total.
One is wrong.
A channel-mapping fix can recover a convincing number while corrupting the evidence underneath. Check every pair before accepting the repair.
Fixed synthetic data · computed in your browser · no key or cloud call
01 / TRY THE REPAIR CHECK
Which repair holds up?
Every pair must agree
READY| Pair | Registered | Candidate | Check |
|---|---|---|---|
| Run a check to compare the evidence. | |||
A matching total is necessary. It is not sufficient.
02 / PRESERVE THE SCIENTIFIC BOUNDARY
The repair cannot approve the claim.
Original claim 3.40 · NOT REPRODUCED
Choose a review intent for this source. Nothing is dispatched, paid for or stored by a service.
Review intent only. No cloud action.
Inspect the proposed source and its bindings
Only fixed parameter arrays differ. Candidate Python is displayed as text and never executed. The source hashes cover protected counts and tests as well as the original analysis.
Original analysis.py
No source loaded.
Candidate analysis.py
Check a repair first.
SOURCE SNAPSHOT SHA-256 —
BROWSER RECEIPT SHA-256 —
03 / SAVED REAL EXECUTION · 26 OCTOBER 2026 UTC
A real model suggestion. A separately verified run.
NVIDIA Nemotron 3 Nano on Nebius Token Factory proposed bounded parameters. A disposable Nebius Sandbox ran the three source variants against unchanged tests.
| Executed variant | S | Checks passed | Observed outcome |
|---|
What this saved run establishes
The prompt supplied the declared A1 channel flip. Nemotron translated that convention into two parameter arrays; it did not autonomously discover the correction. A deterministic adapter built the patch. The wrong-pair control was operator-authored.
The incomplete-evidence browser case was not part of the real three-variant Sandbox run. Twelve checks include three intended control failures. These fixed examples are not an accuracy benchmark or a physical quantum experiment.
Hashes establish content integrity, not a provider signature. Actual billed money and account-wide spending controls remain unknown.
How the browser check works
SHA-256 pins verify the bundled synthetic source files, source snapshot and policy before any result is enabled. The browser recomputes each correlation and the CHSH total with exact rational arithmetic. It supports these three fixed candidates only and does not parse or execute arbitrary Python.
Browser reviews have their own receipt schema. They retain the original Python reference receipt hash for traceability; they do not claim to be a new Python or cloud execution. Review intent remains in this page until you download it. Reloading clears it.
There are no API calls, secrets, uploads, payment actions or live cloud controls in this page. Saved cloud results are playback. Publishing access to this site is a separate deployment step.