Start from the next action a person can take. If the decision is already scoped, a briefing or a review workspace shows the result, the evidence, and the status. If the scope is not known yet, a guided question or an open conversation can narrow it. Those are complementary modes. An empty chat box is a weak default when the organization already knows which decision is under review.
The decision-first article explains how to scope the decision before a model speaks. This article is the interface that scope becomes. It is a conceptual layout, not a screenshot of a released Orisdale workspace, and nothing on this website answers a business question with an agent.
Conceptual interface. It is not a screenshot of a released Orisdale workspace, and this website does not run an agent.
Three modes, one decision
A briefing fits when the audience needs the same packet: the question, the governed result, the period, and what is unresolved. Executive Intelligence is that kind of blueprint. It is one briefing with supporting evidence, not a claim of a deployed executive product.
A guided question fits when one parameter is missing. “Which department?” or “Which period?” is a better prompt than a blank box, because the answers map to fields a calculation can accept. A suggested question may refine scope. It must not imply that submitting it runs an agent on this site.
Open conversation fits when the person is still finding the decision, or when they need to ask what a term means. It is a poor place to hide a consequential action. The Workflow Agent, which remains in development, is an authoring product for a reviewable workflow artifact. It is not this decision workspace, and it does not approve a workflow by itself.
What the surface shows
A usable workspace for a known decision keeps five regions visible without a second click:
- The question and the scope that was accepted.
- The governed result, with its definition version.
- The evidence, including anything still missing.
- The review status, including “not recorded” when that is the truth.
- A follow-up that stays inside the scope, such as another department in the same period.
Finance Intelligence is an illustrative interface for a variance question. The proof explorer is the concrete place on this site where a person can change which synthetic department is in view. That interaction is a case selector over published files. It is not a chat with a model, and it is not a live warehouse.
Missing data without a dead end
If a required field is absent, the workspace should say which field and which source was checked. It should not replace the gap with a plausible sentence. The reviewer can request the field, reject the item, or leave it unresolved. Those are the same review states described in the human-review article. The workspace’s job is to make the state visible next to the result, not to animate it away.
Uncertainty belongs in the same region as the result. “The percentage is unavailable because the budget is zero” is a clearer statement than a badge that only says low confidence. A model score is not a reason to show or hide an action.
Follow-up that does not change the contract
A contextual follow-up can ask for another slice that the same definition allows: a different department, the prior period, or the evidence file. It should not invite the model to invent a new formula. If the person asks for a cause, the workspace can show whether a linked evidence record exists and whether a person has confirmed it. The Finance example keeps human confirmation “not recorded” or “not applicable” exactly as the artifact says.
How to judge the interface
Watch a reviewer who already knows the decision. Can they find the scope, the number, and the blocked action without reading a transcript? Can they tell a briefing from an action? Can they see missing data without opening a debugger? Those are usability questions. This article does not cite an adoption study, a 90-day retention figure, or a claim that conversational analytics has ended.
Chat remains useful. It is the wrong primary surface when the next action is a review of a known result. Put the result on the page. Let conversation sit beside it for the questions the result does not answer.
Limits
The drawing above is conceptual. Orisdale has not released the workspace it sketches. The architecture page remains the reference for how a question becomes an approved tool call and a reviewed result. If you want to walk one existing decision — a variance, a briefing, or an operations exception — start from the contact page.
A path through the workspace
Consider a finance reviewer who already knows the decision: which variances need attention. The workspace opens on the question and the period, not on a greeting. The result region shows the approved budget, the actual, the variance, the direction, and the category from the contract. The evidence region shows whether a driver record is linked. The status region says that human confirmation is not recorded, when that is what the artifact says. A follow-up offers the other synthetic departments in the same period. That is the proof explorer’s job on this site, drawn here as a general pattern.
Nothing in that path requires an open chat. If the reviewer asks what “material” means, a guided question can show the contract thresholds: the absolute amount and the percentage, in the published order. If the reviewer asks why the variance happened, the workspace shows the claim status and stops. It does not invent a cause, and it does not offer a button that posts an adjustment.
A different decision, an executive briefing, uses the same anatomy with a different action. The result is a short brief. The evidence is the supporting notes. The status may be “for reading,” which means there is no blocked transaction. Labeling that status matters. A reviewer who cannot tell reading from approval will either ignore the brief or treat a click as a decision the brief never requested.
Conversation still earns a place beside the workspace. A person who does not yet know whether the question is a variance, a coverage figure, or a weighted pipeline amount should be helped to choose. Those three are different, as the revenue article separates coverage and weighting from predictive accuracy. The chat can ask which one is meant. It should not compute a fourth, unnamed metric in the reply.
The in-development Workflow Agent remains a separate surface. It is for assembling a reviewable workflow artifact from requirements, with manual import, binding, validation, and execution left to the author. Putting that product’s name on a decision workspace would blur an authoring tool with a business review. Keep the link in the product page, not in the briefing layout.
Judge the design with a rehearsal. Give a reviewer the packet and no transcript. Ask them to point at the scope, the number, the missing evidence, and the action that is blocked. If they cannot, the workspace is still a chat with decorations. If they can, you have an interface that matches the decision you already scoped. This site cannot run that rehearsal for you. It can show the regions and the proof explorer, and it can keep the claims inside the evidence you can download.
When you rehearse, include one missing-data case and one case where the evidence is linked but unconfirmed. The reviewer should be able to say which is which without guessing from tone. If the workspace only looks finished when every field is populated, it will hide the situations a real review is for. Leave the empty state designed, labeled, and reachable from the same page as the complete state.