Eve library

Eve interactive answers, live dashboards and editable motion — PRD

PRD

Revision 3

On this page

Status and intent#

Proposed product requirements, prepared 2026-10-09 for Elikem's request to proceed with all recommendations from the Claude Dashboards/Motion and ChatGPT Intelligent UI analysis. This is a delivery proposal awaiting independent approval, not a claim of implementation. Program family: eve-interactive-artifacts. Existing related work is reused through actual task dependencies.

Eve should choose an appropriate working interface for a question, let Elikem inspect its evidence and change its inputs, and preserve useful results as editable artifacts. The same pinned content can become an exported report or short explainer. A simple question should still receive a simple answer.

Scope and authority#

Eve's operator workspace serves Elikem and agent lanes. Existing DS00/DS07 decisions place authoring in the admin app through the BFF; canonical content is stored in Eve. Collaboration means human-agent edits, comments, presence and conflict handling. Public share links, additional human roles and tenant administration are not introduced. The explicitly rejected DS07.2 is not revived. Customer web and supported mobile clients receive the common rendering contract with their own authorized data; they do not acquire operator board access.

Generation and saving a draft do not grant publication, UAT promotion, model spending or credential-administration authority. Existing confirmation and approval boundaries remain. A model provider change is not a prerequisite, and product quality is measured with the approved inexpensive OpenRouter route.

Requirements#

R1 — Adaptive answers. Eve chooses text, metric/table/chart, comparison, timeline, diagram or parameter controls to fit the question. A response can combine text and useful interactive content. Accepted components appear progressively. Unknown or malformed output has a useful fallback. The user can stop the turn and continue or revise the accepted result safely.

R2 — Inspectable results. Every source-derived number carries its metric definition, query/version, normalized filters, observation time, source IDs or revisions, and any truncation/partial-data warning. An observed zero, an empty result, an unavailable source and an estimate are distinct. Do not invent historical comparisons without historical records.

R3 — Live dashboards. Natural-language questions and direct controls can create and revise a persistent dashboard over authorized tasks, reviews, releases and dependency records. Manual refresh is always available; recurring refresh is opt-in. Show last successful refresh per chart. Failed refresh retains a labelled previous snapshot rather than presenting an invented fresh result.

R4 — Artifact lifecycle. Save, reopen, fork, inspect source, compare revisions, make targeted edits and export. Use stable node IDs and immutable source/data snapshots. Data refresh does not silently edit the design or change an already exported artifact. Concurrent conflicting edits are visible and recoverable.

R5 — Scenario tools. Typed inputs and deterministic, versioned calculations drive useful what-if tools. Initial examples are capacity/workload and resource-cost scenarios with explicit assumptions, units, rounding and limitations. Local controls do not make model calls on every change. Do not present a scenario as a validated forecast.

R6 — Editable motion. Turn a pinned dashboard, document or presentation into a short sequence of editable text, charts, shapes and authorized images. Provide storyboard/timeline, play/pause/scrub, timing and direct text edits, conversational revisions, captions/transcript and MP4 export. Generated footage is outside this feature. Narration is optional and uses an already authorized route if available; silent captioned output remains useful.

R7 — Reusable visual systems. Use existing design tokens, templates and reviewed design-system releases for dashboards, scenarios and explainers. Maintain version pins and conformance. Shared-template changes do not silently alter saved artifacts. Preserve token-source and design-system publication decisions.

R8 — Collaboration. Reuse anchored comments, origin attribution, presence, disjoint-patch rebase and explicit conflict handling for Elikem and agent lanes. Preserve revisions and comment history, including orphaned comments after node deletion.

R9 — Inputs and connectors. Support bounded CSV/JSON datasets with explicit schemas and hashes. Bind dashboards through the governed connector registry; first external proof is an authorized read-only OpenRouter usage source. Only documented actual provider fields are supported. Warehouse and CRM adapters get an extension contract, not an unimplemented capability claim or an obligation to buy accounts.

R10 — Exports. Provide data/specification bundles, static formats and editable handoffs through the existing export framework, and MP4/caption output for motion. Preserve source snapshots, fonts, units and supported editability. State unsupported format features explicitly. Export does not create a public link.

R11 — Cross-surface usability. Admin/customer web and supported native assistant surfaces share semantics and clear capability fallbacks. Support keyboard, screen readers, large text, contrast, reduced motion, accessible alternatives to charts and readable mobile layouts. Full motion authoring is admin-first; mobile must support playback, source inspection and the admitted artifact interactions.

R12 — Reliability and controls. Enforce scoped access on every query, refresh, saved artifact, media read and download. Revoke cached and queued access correctly. Bound time, rows, node counts, render duration, CPU/memory and concurrency. Provide cancellation, deduplication, recovery and observability. No success state without an actual result.

Flagship journeys#

J1: Ask what blocks delivery across Oshun; filter by product; inspect a count; select a dependency; open its authoritative task/review/release; save; leave; reopen; refresh after a real underlying change. J2: Fork a capacity scenario; change a declared assumption; compare results; save and reopen the same calculation version; distinguish the model from observed data. J3: Freeze J1's snapshot; produce a 30-second explainer; change a caption and timing; export and decode MP4; verify every number against the selected snapshot. J4: Create from a pinned template; import CSV; bind the admitted connector; accept disjoint agent edits; resolve a conflict; export a reproducible bundle.

Design direction#

Visual thesis: calm, precise working views with strong typographic hierarchy and the data as the visual focus. Content plan: the requested answer and main interaction first, source/assumption inspector second, details and history on demand. Interaction thesis: progressive arrival of usable content, restrained selection/layout changes, and deliberate motion that explains sequence or causality. No marketing hero or ornamental dashboard-card mosaic on operator screens.

Delivery and success#

M1 accepts interactive answers and persistent editing. M2 accepts the live delivery dashboard. M3 accepts dashboard-to-motion and reproducible exports. M4 accepts templates, scenarios, imports/connectors and collaboration. FINAL requires all requirement evidence, independent review, tested main, a passing normal Dev release and operational handoff. UAT promotion remains a separately authorized event.

The delivery specification defines the proposed measurable budgets and acceptance matrix. Product completion requires working flows and exact evidence; schema presence, task count, a demonstration or a historical shipped label does not establish delivery.

Sources and existing work#

Product inspiration: https://claude.com/resources/articles/dashboards-and-motion (2026-10-08); https://openai.com/index/gpt-6-for-everyone/ (2026-10-07). These sources establish feature ideas, not Eve implementation or API availability. Existing approved decision: document 9b0e549a-6453-834e-a597-a15b969cf738 revision 4 (ADR-0083). Existing board requirements: eve-sota-gap-closure:8.5, 8.14, 8.15, 8.19; eve-gap-analysis:DS00, DS01, DS07, DS08, DS12, DS13 and R15. Current code observed at e41d235d4b: generative-ui.ts admits only confirmation; turn-stream-frames.ts supplies shared SSE; PresentationEditor.tsx supplies structured presentation editing. This source inspection is not a deployed capability audit.