Product updates
What has changed for the people building, reviewing and carrying forward AI system evidence.
The system workspace now brings baseline comparisons, evidence traces and reviewer decisions into a more focused interface. Project navigation is simpler, and source details open in context instead of competing with the main task.
System Records, the Gap Workbench and Evidence Pack views put the current state and next action first. Mobile inspectors preserve the same decision context on smaller screens.
Research setup now separates available capabilities from the evidence and review steps needed for a selected experiment. Reviewers can inspect source citations and proposed changes before advancing the bounded execution path.
Research records and inspectors are easier to read on smaller screens, keeping the decision and its supporting evidence together.
Literature evidence retains the source version and supporting excerpt used in review. Experiment comparisons preserve their parent and baseline context, and reviewers inspect proposed code changes before the bounded digit-recognition demonstration runs.
Accepting a research finding remains separate from accepting evidence for a governance pack.
Proposals, literature evidence, experiment lineage and reviewer-led findings connect in the research workspace. Failed literature extraction is distinguished from an empty result, so missing evidence is easier to investigate.
Customer-facing assistants also gain focused transparency actions and a draft Transparency Pack for Article 50 review.
The project Research Workspace and organisation Research Library connect datasets, repositories, experiments, findings and decisions. Task packages and result imports support the digit-recognition demonstration without implying general autonomous execution.
Evidence Runs validate dataset readiness before starting. Completed work stays inside the selected AI system. System, dataset and pack views make blockers and next actions clearer, alongside published private-beta data boundaries.