Follow your curiosity through a city of signals, learning, poetry, and proof. Every discovery has another way in.
Change a signal. Make a prediction. Watch what remains. Then inspect the evidence.
Seven spectral sectors. Two retained. Change the initial overlap and see how the filter reshapes the signal.
| Sector | Dimension | Population | Selection |
|---|
Four graph-derived quantum gates change a fixed retained input while preserving its norm. Move an angle and watch the observable respond. These controls are separate from the spectral filter above.
| Coordinate | Population |
|---|
The filter leaves the J = 2 and J = 5 sectors unchanged and contracts the other sectors. The complement norm is bounded by (15/17)n times its initial norm. Conditional fidelity is undefined when the selected overlap is zero.
A small neural network learns from synthetic points. Predict, train, compare—and keep neural accuracy separate from spectral projection.
This browser lab runs deterministic gradient training on synthetic data. Its training accuracy does not measure performance on unseen real-world data. The existing ENTRAIN instrument includes an additional feature encoding and E47 projection monitor; the Signal Garden here is a separate experiment.
| Point | X | Y | Label | Probability | Prediction |
|---|
Explore the existing surfaces. Each destination keeps its own purpose and evidence boundary.
Inputs, assumptions, sources, and limits stay attached to each experiment.
H = V₂ ⊗ V₂ ⊗ V₂, dimension 125.
K = (C − 6I)(C − 30I).
ker K has dimension 47.
Γ = I − K²/99144.
The browser Signal Garden computes seven-sector norm observables. It does not reconstruct phases, compile quantum gates, or run on quantum hardware.
Inspect the matrix proof ↗Open the research workbench ↗A two-input, one-hidden-layer neural network with a scalar probability output. Deterministic datasets and initialization make experiments replayable.
Training uses synthetic labels. Accuracy and loss are training metrics. A low loss is not evidence of quantum advantage or a physical theory.
Read the executable model ↗Inspect numerical checks ↗Replay files contain validated inputs. Imported results are recomputed. Saved progress stays on this browser unless you export it.
Stable identifiers, explicit scope, versioned inputs, and attributable sources. Machines can inspect the same experiment without interpreting the visual scene.
Machine contract ↗Instrument catalog ↗Public evidence snapshot ↗Embedding contract ↗Choose your own date and save a 30-minute session with your current replay link. Calendar changes happen only when you confirm in your calendar app.
A curated snapshot of explicitly public registry records. Registration is distinct from independent verification.
Published snapshot · 13 October 2026