Shipped
- A task-led learning hub now links stable workbench routes to their evidence, limits, owners, and maintenance checks.
- The research entry page now explains the first bounded outcome, provenance expectations, privacy boundary, and execution limits before a visitor opens the workspace.
- Activation measurement is optional, browser-local, category-only, and consent-gated. It excludes model values, sources, prompts, diagnostics, and identifiers.
- Shared workbench controls now provide consistent labelled, keyboard-reachable states across the surfaces that use them.
Known limits
- SBML execution remains bounded by the current format and operation capability report. Unsupported constructs stay visible and can block a run.
- CellML, NeuroML/LEMS, and PBPK documentation names source-bound inspection and planning boundaries. It does not claim a selected execution backend where none exists.
- A completed run, export, assistant explanation, or parser check does not establish a biological, clinical, or scientific conclusion.
Next work
Future notes will describe a change only after it lands with its route, evidence contract, and known limits. Case studies will be published only with attributable user evidence.