What was run, when, and on what.
This page is generated from the reports of real test runs: unit tests against an independent reference, end-to-end checks in a real browser, and live runs of the on-device model. Anything that has not been run is shown as not run.
Mathematics, checked independently.
The circuit is compared with a separate numpy dense-matrix implementation, to within 1e-10. The exact checker is compared with a dynamic-programming knapsack solver on 600 random inputs. Malformed input, ties, zero budgets and impossible jobs all have their own cases.
| Test file | Passed | Failed |
|---|
The real site, driven in a real browser.
The built site is served with its production security headers and driven end to end: the home page, the lab, share links, export and replay, malformed input, phone width, and a check that nothing contacts a third party before you opt in to the AI.
| Check | Result |
|---|
Live inference, actually run.
Each row is a real run of the optional language model in a browser. The model loads, drafts a job table, the planner runs on the confirmed draft, and the model rewords the result under sentence-by-sentence checks.
| Date | Browser | GPU | Load | Downloaded | Draft | Explanation | Result |
|---|
Not run: Firefox, Safari, mobile browsers, other GPUs, macOS and Linux. A CPU (WebAssembly) fallback was tried and does not work with this model.
Replay any exported run.
Drop a run file exported from the lab. Your browser re-runs the simulation with the same inputs, seed, shots and layers, then compares the angles, every sample count and the candidate with the file. Nothing is uploaded.
Drop a qubitflow-run-*.json file here, or .