Your program. Your reusable interface.
Take a scientific program beyond the built-in workflow catalogue. The modular r34 framework turns compatible Python programs into private, reusable forms—with AI-assisted authoring, explicit validation and retained results.
From a custom Qiskit model to a private workflow
Start with a six-qubit kicked-Ising Floquet chain: local phase-flip noise, an ideal inverse echo and quantum-state diagnostics. There is no built-in registered workflow for this exact experiment; the integrated Qiskit SDK can express it. Follow its program into the assistant, watch the generated GUI settings change, and inspect new calculations, private versions and results. The walkthrough also connects the interface to B2 project authorization and B3 AI disclosure enforcement.
Private workflows: GUI and server walkthrough · Narrated · English captions
Learn why the Python program is needed, how its model and outputs are built, and how to author it with the workflow assistant or the development guide. Then follow the custom Qiskit study through validation, reusable controls, private versions and results, with B2/B3 server checks. Real calculations use the designated developer executor; AI replies are scripted. Production container isolation is outside this recording’s scope.
The study bundle includes the supplied program, V1 and V2 workflow definitions, modular source, recorded settings/results and an independent numerical verifier. These are local model calculations, with their assumptions and scope documented in the included README.
What the Python program does
The program implements your model: it reads the chosen settings, constructs the initial state, performs the numerical evolution and computes the observables. The workflow definition connects that calculation to reusable controls, checks and result views. For the Floquet study, the source specifies the Ising gates, transverse rotations, phase-flip channel and quantum-state diagnostics.
Attach the complete Python source and explain its inputs, units, outputs and dependencies. Ask the assistant to adapt it to the workflow contract while preserving the intended mathematics. Review the proposed files and validation cases before running them.
Use the Polynomial starter as a contract example, then ask the workflow assistant to implement your stated equations, conventions, initial state and observables. Require complete source, explicit assumptions and meaningful checks. Review and refine its proposal; a generated program still needs scientific and numerical validation.
The development guide walks through both routes and includes a complete six-to-eight-qubit Floquet example. Large quantum states are written to downloadable artifacts, with compact results for the GUI.
The end-user development path
Open My workflows → Create with AI. Attach or paste an existing Python program, or ask the assistant to build one from your mathematical specification. Define the model, units, initial and boundary conditions, observables and justified checks. No completed simulation is required.
The assistant prepares a private bundle with modular source files, typed inputs and structured outputs. Review the source and controls before executing it. Manual bundle import is also available.
Run the declared cases, inspect their real calculations and expectation results, then explicitly activate the passing version. The assistant cannot activate it on your behalf.
Edit the saved GUI controls, run the retained program and inspect metrics, tables, plots and downloadable artifacts. Subsequent runs do not require another AI call.
Return in later sessions, inspect run history, reuse settings and export a bundle. Source or schema edits create a new version that needs validation and activation. Use the user controls to archive, restore or delete a workflow; saved assistant conversations have their own Delete control.
Return to your saved work
The shared standalone and SaaS Assistant opens saved conversation information without automatically importing a completed simulation. Select an existing conversation to continue it, use Open selected run when you want to capture that run’s evidence, or explicitly create an authoring conversation for a new workflow. Existing attachments and personal provider settings stay part of their saved research context.
One modular framework
The Python function and script adapters accept compatible programs using provisioned scientific libraries, including combinations of APIs. Typed schemas define number, integer, choice, Boolean, text, JSON and file inputs. Structured outputs drive metrics, tables, series and artifacts.
Your bundle holds its own program files and form definition. The shared installation supplies the generic framework; adding a personal workflow does not rewrite the installed scientific core. Existing private source layers remain a separate mechanism.
API-agnostic does not mean every library, remote API or hardware target is immediately available. Required versions must be provisioned in the isolated runtime. The shipped private executor has networking disabled; remote services, hardware access and additional runtimes need separate operator support.
Explore the installed integration routesFrom workflow results to evidence
Declare the meaning of retained outputs so the shared reporting framework can check eligible methods, preserve assumptions and evidence, and produce linked figures and presentations. Scientific calculation, report regeneration and selected-view export remain separate actions.
Explore the Intelligent Scientific Analysis and Evidence SystemPrivate state and enforced boundaries
Hosted workflow operations retain the user and project context. Permission checks apply on the server, including direct requests and access changes during an operation. Sharing a project does not automatically publish one user’s private workflows to another member.
Read project authorization rulesAI authoring obeys effective organization and project policy, approved providers, classification and source-disclosure rules. Required consent is checked before provider requests, including retries. A private workflow remains usable without new model calls after activation.
Read AI policy and consent rulesVersions bind the program, declared dependencies, source profile and runtime identity. Run records retain inputs, outputs and execution evidence. Archiving removes a workflow from ordinary selection while retaining its history. Show archived workflows to restore one, or use the explicit Delete control for permanent removal. Export/import remains available for portable copies. These lifecycle decisions belong to the user, not the assistant.
Understand storage and confidentiality boundariesComplete end-user development guide
The current 50-page guide covers writing a program from a mathematical specification, adapting existing source, both Python adapters, typed controls, files, results, checks, dependencies and the complete lifecycle. Its worked Floquet study includes the authoring prompt, scientific conventions and exact JSON limits: 2 MiB, 100,000 value nodes and depth 32. Seven importable examples are included in the application package.
Delivered r34 package
The cumulative record includes checks of the framework, real numerical programs, browser workflows, authorization, AI policy and release integrity. Repeated identities count once. The latest Assistant consolidation passed 305 targeted Python tests and 232 UI fixture cases across 14 commands. It brings the approved recovery, saved-conversation loading and opening fixes into the shared standalone/SaaS release. Subtests and assertion checks remain separate; no simulation-result cache is introduced. Autonomous investigations and upload maintenance are also included in the ledger; these releases did not rerun the full scientific benchmark campaign. The designated developer executor ran the retained numerical checks; target-host deployment and container isolation need their own qualification.