ruQu ships an offline laboratory for observer consistency experiments

What this is about

ruQu is a quantum simulator written in Rust and compiled for the browser and the command line. Simulators make experiments cheap, which is useful and also dangerous: an experiment that is cheap to run is cheap to run badly, and a result with no record of how it was produced is not a result.

This change is less about quantum mechanics than about what it takes for a simulated experiment to be worth quoting.

What changed

Pull request #4 was merged on September 20 as commit eb127994. It adds an offline, reproducible observer consistency laboratory that runs against the bundled simulator, with an optional adapter onto pinned builds of two sibling projects.

The mechanics, as described in the pull request:

• A command with a run count and a seed, plus a benchmark mode, so a run is repeatable by anyone with the same two numbers.
• Bell and CHSH protocols alongside local deterministic and dephased controls, so a result has something to be compared against.
• A canonical hash evidence chain binding the protocol to the declared simulator build, checked before anything is created rather than after.
• Real field events from the synthetic simulator with privacy and provenance checks, and an explicit rule that a run claiming to replay captured data rejects synthetic evidence.
• Actual graph nodes and relations for rooms, events and semantic state, with the toy geography declared as unregistered rather than passed off as real.
• Bounded inputs, capped at 32 MiB and ten thousand rows, with pinned dependency revisions and a committed lock file.

The verification recorded in the pull request is specific: 1,376 tests passing in the existing Rust workspace, 40 observer suites, 12 native integration tests, strict lint and formatting, and an end-to-end run accepting four workloads up to ten thousand rows while rejecting fourteen malformed cases with no partial output.

Get started

Prerequisites: a Rust toolchain, Node.js, and a clone of the repository. The laboratory is offline by design, so nothing needs network access or an account.

observer-lab --count 100 --seed 42

Expected result: a completed run of the stated size with an evidence chain bound to the declared simulator build. Changing the seed changes the run; keeping it should reproduce the same one. Build the command from the repository first, following the project documentation, since there is no published package for it.

The command is quoted from the pull request. It was not executed here, and the figures above are the maintainers' recorded results rather than measurements taken for this article.

Use it today

Practical case: you want to show a colleague that an observer effect in your setup is real rather than an artefact of how you ran it. Input is a seed and a run count. The workflow is to run the protocol and its deterministic control with the same seed. Output is two evidence chains that can be compared independently.

Reader acceptance test: run twice with the same seed and confirm the evidence chains match. Then change the declared simulator build and confirm the run refuses rather than proceeding with a mismatched binding.

Push it further

Experimental commentary, and this story is filed as experimental deliberately. The transferable idea is the refusal rule: evidence that would be synthetic in a run claiming to be captured data is rejected outright. Most experimental harnesses would warn and continue, which is how a synthetic number ends up in a slide six months later.

Limitation: everything here is simulated. A hash chain proves a run happened the way it says it did, and proves nothing about physical reality. The native adapter is optional and depends on pinned builds of two other projects, so it will not necessarily build for you. Falsifiable test: feed it an input over the documented bounds and confirm it rejects the whole run rather than emitting a partial result.