A task is not one call to a model. It moves through a fixed pipeline of stages, and build scope decides which of them run. This page is the full list. The user-level account is on How intelligent.page works.
The stages
Interface Selection, Component Planning and Component Props Compilation always run — without them there is no answer. Everything else is what scope turns on, and the exact matrix is on what runs at each build scope.
Operate, Apply and Read Back are one pass and cannot be half-enabled — a refinement you apply without re-reading is worse than not refining at all.
Where the human gate sits
Ask for permission at interface selection (directory settings) pauses the run between Interface Selection and the Gate, showing the surfaces it intends to build. The "Continue, don't ask again" action on that pause is the same switch, not a one-off. See Interface selection and the permission gate.
The organisation stages
The pipeline above is how a task runs. The organisation model it reads and writes is organised along a different axis — the stages of a business — and the company audit reads the website along them, in this order:
Market · Customer research / segments · Customer persona · Day-in-the-life · Customer journey · Micro-moments · Problem validation · Value propositions · Product · Brand · Pricing · Channels / moment-to-channel mapping · Business model · Strategy · Objectives, metrics, evidence · Roadmap and decomposition · Team and capabilities · Processes and systems · Resources / budgets · Operations · Acquisition and revenue · Customer / business outcomes · Visibility and governance · Feedback / experiments · Meetings
Each stage has a default surface — a persona is cards, a journey is a flow diagram, a roadmap is a timeline — which is what Interface Selection starts from when a request lands on that stage. On Connected, work on one stage triggers the stages that depend on it; on Entire system, every stage is rebuilt whether triggered or not.
Agents
Within a stage, work can fan out across several agents so that depth on one part does not hold up the rest. How many is a function of your plan — see Plans.
Personalities
On some plans you can set the voice and stance each stage writes in — so a review stage can be made more sceptical than the default.
Why a review stage exists
Long pieces of work drift: a conclusion in one artefact quietly contradicts another. The Reviewer exists to catch that, and what it finds becomes issues rather than silent edits.
Following along
Activity shows the stages of a real run, in order, with what each one read.