# Step Classification and Slice Selection — Stage 3 of the [method](https://aimasterz.com/docs/method). Classify every step from every process card, then choose the first slice. ## Executor rules | Executor | Use for | | --- | --- | | **Code** (deterministic) | Rules, routing, calculations, record keeping, deadlines, retries, data movement | | **Agent** | Interpretation, research synthesis, drafting, classification with ambiguity, summarizing | | **Human** | Decisions with accountability, approvals, relationships, anything the customer says stays human | ## Risk classes | Field | Values | | --- | --- | | Action | `read` (observe), `draft` (produce something not yet live), `act` (changes something real) | | Reversible | yes / with effort / no | | Outward-facing | yes (customers, public, partners see it) / no | | Money | yes / no | | Data classes | public / internal / confidential / restricted (security baseline) | **Approval rule:** any step that is `act` and (outward-facing, irreversible, or money) needs exact human approval. Read and draft steps run without approval when their grants allow it. ## Classification table | Process | Step # | Step | Executor | Action | Reversible | Outward | Money | Data class | Approval needed | | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | ## Process scoring Score 1–5 each. Priority = (Value × Feasibility) ÷ Risk. | Process | Value (time saved, metric impact) | Feasibility (data + connections exist) | Risk (blast radius) | Priority | Notes | | --- | --- | --- | --- | --- | --- | ## First slice | Field | Value | | --- | --- | | Process(es) | | | Why this first | | | Steps automated vs kept human | | | What is measured | | | What is explicitly out of the slice | | ## Gate - [ ] Every step has an executor and a risk class. - [ ] Every `act` step that is outward-facing, irreversible or involves money has a human approval step. - [ ] The first slice is chosen with a written reason.