​
The four jobs

Ask the model on the left starts one: choose the job, an outcome for a playbook draft, extra pages to read, and optionally pasted terms for pages that block fetching. Results appear on the right within a minute or so. Without a model configured only the scan is available; the panel says where to set the endpoint.

​
Research cards

Each card shows the job, whether it ran automatically (from the new-provider pipeline), its tokens, cost and time, the model, and its status: queued, running, ready to review, applied, or failed with the error. The result depends on the job: the scan lists the pages it chose and the candidates it scored; an analysis shows a confidence badge, the proposed outcomes and documents, the summary, the policy text, citations and open questions; a playbook draft shows the steps with any validation problem and the assumptions to check; a change review says whether the policy changed and recommends an action.

​
Applying

The button under a ready result is the only way a model’s output reaches the record: Add these pages to the provider, Apply to the provider, Save as a draft playbook or Acknowledge. Applying an analysis un-verifies the provider so a person reads it before publishing; applying a playbook draft saves it as a draft on the Playbooks tab. Every model call is logged with tokens, latency and cost under Settings → LLM. The text of each prompt, and the placeholders it is given, can be read and edited by an admin under Settings → Prompts.