Reproducible examples
Prompto’s browser demo uses synthetic fixtures. Its sample outcomes illustrate the workflow; they are not a measured research dataset.
Open the demoFrom questions
to evidence.
Investigating AI integrity, deployed systems and the physical world they operate in.
Projects
Methods, existing evidence, and what comes next.
01
Vraelis Prompto
02What changed in a model folder?
File identity, structural findings, and the distinction between recording a baseline and verifying it.
Implemented workflowVraelis Contour
03Approval and reported runtime identity
A release can be approved while the runtime still reports an earlier release. Preserve both facts.
Recorded reference experimentVraelis Imprint
04Connecting a component to its evidence
Geometry and motion provide context. Component identity and source records make that context traceable.
Visual studies / proposed methodAcross the research program
Testing a specific claim
Make the test conditions, evidence and failure criteria explicit before reporting a result.
Evaluation protocolResearch areas
The questions behind the projects.
Model integrity
Whether the model that runs is the model that was approved.
Adversarial threats
Inputs, updates and supply paths built to change what a model does.
Machine trust
Explicit identity, limited permissions and checks at execution boundaries.
Security evidence
Records a person can read, kept in order, including what could not be confirmed.
Give every component
a traceable record.
Explore the proposed method for linking geometry, revisions and motion to their sources.
Read the projectData & references
Start with the source and its scope.
Recorded experiment
The Contour documentation explains the controlled CPU experiment and its historical test results. Follow the record’s version and limits.
Read the evaluationSource library
Each project links its relevant documentation and external references. Imprint visual sources are cataloged separately from measured evidence.
Visual attribution