Design

Single-pass mapping engine: translate each Heroku resource to its AWS equivalent using deterministic lookup tables. No clustering, no dependency graphs — resources are processed as a flat list in input order.

Composed of a mapping fragment + one assembler (declared in the frontmatter _fragments/_assemble); the interpreter runs the fragments whose _trigger is true, then the assembler, and evaluates the _knowledge _when guards to load only the sizing tables the inventory needs (see INTERPRETER.md § _knowledge). The compute target branch is selected by design_constraints.compute_target: Elastic Beanstalk is the default, Fargate is the direct-container override, and EKS is the Kubernetes-team alternative. The preferred shape is compute_target.default plus per-formation overrides[]; legacy compute_target.value is accepted only for reused older preferences. The EKS branch (design-eks, fired by its _when trigger when the global compute target selects EKS) is an ALTERNATIVE path, not an addition: it maps ALL formations to EKS pods + an eks_cluster aggregate instead of the EB/Fargate paths. On the EB default path, persistent non-web formations with quantity > 1 route to Fargate to preserve Heroku worker capacity because EB SingleInstance cannot horizontally scale that process model.

Scope Boundary

This phase covers Heroku → AWS Design ONLY.

FORBIDDEN — Do NOT include ANY of:

• Cost estimates or pricing calculations (that is Phase 4 — Estimate)
• Terraform generation or HCL code (that is Phase 5 — Generate)
• Migration scripts or runbooks (that is Phase 5 — Generate)
• Data migration procedures (that is Phase 5 — Generate)
• Feedback collection or plan sharing (that is Phase 6 — Feedback)
• Clarify questions or preference gathering (that is Phase 2 — Clarify)
• Resource discovery or API calls (that is Phase 1 — Discover)

Your ONLY job: Map each Heroku resource to its AWS equivalent. Nothing else.