Follow the r34 SaaS workflow and examine how private layers can override equation implementations, with a recorded Bell study and retained equation-level test evidence. Use the chapters to find a step, or read the complete transcript.
Inside the research workbench5:35
Recorded r34 SaaS walkthrough · Narrated · English captions
Actual r34 interface and retained equation-override test evidence, with scripted AI responses and real numerical execution. Storage setup is shown.
Run the PF application in your own environment. Install the dependencies for your selected scientific workflows, configure your AI credentials and open the local research workbench.
Use an institution- or operator-managed PF deployment. Enable the AI assistant, configure approved provider access and choose managed storage or an optional personal destination.
You need access to a PF Simulation application package or an existing hosted workbench. This public site provides documentation and examples; it does not run the research application in your browser.
Guides from the r34 package
From the first question to the next private layer.
Select a guide to read it here. Each guide includes a table of contents, a formatted download and a print-friendly view.
Start with the integration catalogue to understand applicability and required runtimes. Use the new satellite and Aer Gallery bundles to inspect saved inputs, numerical outputs and exact reproduction steps.
The 223-page core scientific manual covers established workflows, configuration, interpretation and reports. Its edition predates the r34 AI, BYOK and BYOS additions; use the current guides above for those features.
Core manual edition 1.17.0 · r26 · 20 September 2026
Corrected r34 release
Validation, with its scope visible.
1,806regression tests
454successful subtests
252interface checks
25numerical lifecycles
All planned checks in the retained corrected-r34 continuation, BYOK and BYOS validation passed with no failures, errors or skips. The final package passed fresh-extraction verification, and 117 audited scientific and supporting files matched the original r34 baseline.
Provider responses were scripted and storage endpoints controlled. Numerical calculations ran in real local subprocesses. This validates the covered application workflows; it does not establish live model accuracy, production load capacity, customer cloud behavior or experimental physics claims.