How the discovery works

There is no fixed list of questions. Before it asks anything, Citeaxis reads every link in your brief (the page text, desktop and phone screenshots, the fonts, sizes and colours measured on the page, the internal pages linked from the navigation) and researches the products people already use for this kind of work. Only then does the discovery agent write the questions for your project.

#What a question looks like

Each one carries:

• the question, in the nouns of your brief (“the viewer”, “the console”, “a room”), never a generic label;
• why it is being asked: what changes in the scenario, the stack or the screens depending on the answer;
• two to five options written for this project, plus “something else” and “skip”;
• sometimes a prefilled answer the agent already read in the brief or a reference; you only confirm or correct it.

The agent never asks what the brief or the references already settle. If the reference shows a dark, dense console, it does not ask about the look. If the brief says AWS, it does not ask where the infrastructure runs.

#Rounds

Usually one round, sometimes two or three (DISCOVERY_MAX_ROUNDS). Each round is a card in the chat; the answers you give become facts with an id the Architect cites when it decides the stack (fact:<id>). Facts the agent settled itself from the brief or a reference are recorded with their source.

#The summary

When nothing left would change the project, the agent closes the discovery with a summary card: objective, who uses it, the flows, the screens per surface, inputs and outputs, integrations, infrastructure (cloud, data sensitivity, operator, monthly budget, availability, load), security, acceptance criteria, assumptions and what is out of scope, plus which references drive the look and which drive the behaviour.

You either confirm it, which starts the complete first version and the proposal, or ask for a change, which re-opens the discovery with your request.