Validation status#

What has been checked, how, and what has not. This page exists so that nobody has to guess.

What is validated#

Claim

How it is checked

The code computes what the equations on the mathematics pages say

tests/test_math_docs.py checks the key equations numerically against the implementation

Every model is fitted and compared the same way

unit tests (pytest), 80% coverage threshold in CI

Docstring examples are correct

doctests in CI

Tutorials run and their stated claims hold

every tutorial is executed in CI; assertions check each claim made in the text

Every public model has a block diagram

tests/test_docs_coverage.py checks the atlas against the package

Circuits match their analytic models

tests compare circuit outputs with the model predictions

Simulator gradients are exact

gradient tests against finite differences

Quantum-like models reproduce published aggregate effects

the Clinton-Gore order effect in quantum_mind.data

What is not validated#

Gap

Consequence

No data from human-robot interaction studies

the robotics results (ask or act, question planning, hand-over, fusion) are simulations with illustrative parameters

Simulated people often follow the same model the robot uses

results such as the hand-over cost comparison favour the order-aware policy by construction

No runs on quantum hardware are included

hardware support is tested with simulators and fake back ends only; the QAOA study uses an exact simulator

The quantum-sensing study uses illustrative sensor numbers

the atom-interferometer noise comes from the quantum limit, its bias walk is an assumption, and no measured sensor data are used

QAOA is compared on problems small enough to enumerate

nothing is learned about problems too large to solve exactly

The ROS 2 node has not been run inside a ROS 2 installation in CI

its callbacks are tested with stand-in modules

References were compiled by hand

the reference-check workflow (below) compares them with Crossref; see its report before relying on a citation detail

How to close the main gap: a pilot study#

A small within-subject study is enough to test the central robotics claim, that asking about trust changes trust, and that question order changes answers in human-robot interaction.

  1. Participants. 40 or more adults; ethics approval and informed consent first.

  2. Task. A robot (or a video of one) hands over six objects with a fixed sequence of successes and failures (applications.robotics.TRUST_PROTOCOL).

  3. Conditions. Ask “Do you trust the robot?” only at the end, or also half-way (between subjects); ask the two clarification questions of HRI_DOMAINS in both orders (counterbalanced).

  4. Analysis. Fit OpenSystemBelief and MarkovBelief to the trust answers and the order models to the clarification answers with quantum_mind.compare; run qq_test; report the question effect with a confidence interval. Use quantum_mind.core.recovery beforehand to check that the planned sample size can tell the models apart (tutorial 2 shows how).

  5. Report. Whatever the result, including a null result for the quantum-like model.

Data from such a study (anonymised aggregates) can be added to quantum_mind.data with a citation.

Reference check#

The workflow .github/workflows/references.yml extracts every reference from the documentation, looks each one up in Crossref and uploads a report with the best match, its DOI and a similarity score. Run it from the Actions tab (it needs internet access, which the documentation build does not).