Acceptance criteria

The eighteen checks that define when Phase 1 is complete, and how Olga's voice is verified.

Phase 1

This is part of what RunOlga delivers in Phase 1.

These criteria define done — for the whole of Phase 1, not just Olga. They turn the promises of the platform, the four capabilities, and the guardrails into checks someone can verify against the founding organization's live instance.

All eighteen, or not done

Phase 1 is not complete until all eighteen criteria are verifiably met. Each is a check, not an aspiration.

Basic functionality

The founding organization is live. Set up as a working organization with real content loaded — mission, vision, values, seats, initial metrics, initial priorities, initial processes.
The leadership team is in. Each person can log in and has the right level of access for their role.
All six main sections work. Leaders can add, view, edit, and track data in Vision, People, Data (Scorecard), Issues, Processes, and Traction (Priorities and Meetings).
Home page and Planner work. My View shows personalized content; the Planner shows scheduled work in day and week views.
All the other tools work. To-Dos, Announcements, Knowledge, and Directory.
All settings screens work. Add Teammates, Feedback, Help, and Profile — see Settings.

Olga working

Voice and speed. Olga responds to Phase 1 questions with her defined personality, in under 3 seconds for simple queries. See Olga's voice.
Conversational data entry works for Priorities, Issues, Processes, and Metrics. See conversational data entry.
Basic questions work. Olga accurately answers "what's happening" questions from the organization's data. See basic retrieval.
Setup clarification works. Olga notices incomplete entries during setup. See setup clarification.
Priority-to-task with scheduling works end-to-end. Create a Priority, Olga proposes tasks with days and times, the user approves, and the tasks appear in the Planner and on assigned seats' To-Do lists. See priority-to-task decomposition.

Behind the scenes

Organizations stay separate. A user in the founding organization cannot access another organization's data through any means — including through Olga. See permissions and isolation.
Every Olga answer is traceable. Provenance logs show exactly where each piece of information came from. See grounding and confirmation.
Permissions are respected. Olga doesn't share what a user can't see, verified with at least three different role combinations across the default roles.
Speed goals are met. Simple queries respond in under 2 seconds, multi-source questions in under 4 seconds, measured on production infrastructure under realistic load.
Works on desktop and mobile web. Laptops from 13-inch screens up, plus phones and tablets through the browser — no downloadable app.
Voice review passed. The founder reviews Olga's actual responses across a representative session and confirms her voice is right before go-live.

Pulse meeting rhythm

A weekly pulse meeting can be run inside RunOlga. The team meets, reviews the scorecard, works through issues, updates priorities, and captures decisions and action items — all in the meeting record.

How the eighteen group together

Group	Criteria	What it confirms
Basic functionality	1–6	Every screen in the navigation is real: the six sections, My View, the Planner, the tools, and settings all work with real content.
Olga working	7–11	The four capabilities function end-to-end against real organizational data, in Olga's voice, at speed.
Behind the scenes	12–17	Isolation, provenance, permissions, latency, responsiveness, and human voice review all hold.
Pulse meeting rhythm	18	The team's weekly operating rhythm actually runs inside the product.

Verifying Olga's voice

Criteria 7 and 17 are judged against the canonical voice artifacts, not a written rubric alone. The Olga System Prompt and Sample Interactions remain the source of truth for how she sounds; verification means reading her actual responses against them.

Sample-interaction fidelity. Olga responds to the canonical Sample Interactions (Part 1, Examples 1.1–1.10) with responses substantially matching the targets — same structure, same tone, same conciseness, no anti-pattern violations. See Olga's voice.
Edge-case handling. Olga handles the edge-case examples (Part 2, Examples 2.1–2.12) appropriately — declining out-of-scope requests honestly, respecting permissions, and accepting user judgment without defensiveness.
Conversation continuity. A user can navigate between sections without losing conversation state; Olga's context updates with each navigation. See how Olga works.
Role coverage. Permission checks (criterion 14) should exercise each of the four default roles — Owner, Administrator, Leadership Team Member, and Viewer — not just three arbitrary combinations.