Eevo

Use this when a user or an authorised agent needs to compare 2-8 concrete options whose costs, benefits, risks or competing interests require judgement, such as comparing suppliers, choosing between project or staffing plans, deciding whether to launch, pilot or delay, or checking an agent's proposed action against its alternatives. The call sends the decision to Eevo, Upright's consequential decision model, which weighs the options separately for each stakeholder. For each stakeholder the result gives the option with the highest expected value and each option's probability of being the best; in Default mode also each option's expected value with a p10-p90 range and the size of each consequence. It says whether the stakeholders share a best option; there is no combined result across stakeholders. Inputs: - context: the situation, constraints, amounts and timing. Eevo has no other source of facts. - options: 2-8, with ids A-H. Include "wait" or "do nothing" as an option when it is a real option. In Default mode (the default) each option lists its consequences: one effect each, described as a change from the current situation (for a do-nothing option, roughly against the others on average) with its amount, a label of at most 60 characters, and whether it is a benefit or a harm. Eevo sizes them against each other and counts an effect listed twice only once. Default mode gives the better answer, since Eevo weighs consequences that are named and sized rather than inferring them, and its card shows which consequences carry each option. Mode "direct" takes options without consequences: a quicker answer with less detail, for when the user asks for one or the consequences cannot be named even roughly. - stakeholders: 1-8, typically the organization, its affected customers and team members, and the environment when the options differ environmentally. - revises and revision_note link a corrected call to the decision it corrects; share_with_org sets whether the user's Eevo organization sees the decision; metadata holds references (issue, pull request, conversation, links) shown in the decision log. Each call is stored as one decision in the user's Eevo decision log. Its id (dec_…) is what eevo_show_decision (the interactive card), eevo_report_chosen_decision, eevo_assess and eevo_share_decision take. A Default mode decision still running after about 40 seconds returns status "running" with its id, and eevo_show_decision returns the result once it is done.

eevo_decide

Use this when a situation needs typed judgements rather than a choice between options: a label from a fixed set, a level on an ordered scale, or the probability that a statement holds (for example how severe, how reversible, or which category). The call sends a state (text or a JSON object) and 1-32 questions, keyed by ids, to Eevo, Upright's consequential decision model, which answers each in one pass: - {"type": "choice", "criteria": {"<label>": "<what it means>", …}} (2-64 labels) returns the chosen label with a probability per label; - {"type": "score", "criteria": ["<level 0>", "<level 1>", …]} (2-64 ordered levels) returns the expected level, with a legend mapping level indices back to the texts; - {"type": "noul"} returns a single probability. Each question may carry instructions. Eevo answers only from the given state. Each call is stored in the user's Eevo decision log; metadata holds references shown there.

eevo_classify

Adds an assessment of an Eevo decision's answer to the user's Eevo decision log, separate from what was done (eevo_report_chosen_decision): overall agreement (agree, partly or disagree) and, for each stakeholder of the decision, whether Eevo's best option for that stakeholder is right in the assessor's view (yes, no or unsure), the option the assessor would choose, a confidence from 0 to 1, and a reason (required for no). assessor names who assessed; on_behalf_of names the person an agent acts for when that is not the signed-in user. Each call adds an assessment; none replaces another. Takes the decision id that eevo_decide returned.

eevo_assess

Records what was actually done after an Eevo decision, in the user's Eevo decision log: chosen_option (the option carried out, A-H, or "none"), decided_by ("agent" or "human") and an optional note. Eevo records for each stakeholder whether the chosen option was that stakeholder's best. followed (true or false) without chosen_option applies only when the stakeholders shared a best option. A later report replaces the earlier one. Takes the decision id that eevo_decide or eevo_classify returned.

eevo_report_chosen_decision

Sets who sees an Eevo decision, for when the user asks to share it or keep it private: shared true makes it visible to everyone in the user's Eevo organization, false to the user only. It applies to every revision of the decision, and only the person who made the decision can change it. Takes the decision id that eevo_decide returned; it does not run the model.

eevo_share_decision

Show the user an Eevo decision as an interactive card: one stakeholder at a time with a picker, each option's consequences (Default mode) or its probability of being the best (Direct mode), and a link to the decision in the Eevo console. Clients that show MCP Apps render the card; others get a short text. Call it once per decision the user should see, with the FINAL decision id: after any revisions (the latest revision's id), not for each eevo_decide call. It reads only your organization's decisions and does not run the model. If the decision's content is not stored (its key's retention is "Not stored", or the retention period has ended) or the decision failed, it returns dataAvailable: false with a suggestion and shows no card; present the eevo_decide result as text instead.

eevo_show_decision