Tiering

gcp-to-aws computes creation_order_depth by walking the dependency graph. Two reasons that is the wrong tool here:

1. Generate wants a cutover order, and the order is always the same. Network, identity, and secrets have to exist before data, data before compute, compute before the edge that fronts it. A computed depth reproduces that order on a complete graph and produces something arbitrary on an incomplete one.
2. Depth is unstable under partial discovery, which is the normal case here. Discover over Terraform alone, or over a repo with one unresolvable module, yields a graph missing edges — and a missing edge silently changes every downstream depth. A fixed tier derived from resource TYPE cannot drift that way.

The four tiers

Assign in order; the first match wins.

compute deliberately outranks data. A merged app+database cluster is a compute cluster: the compute is what gets sized, what dominates the estimate, and what the report leads with. Calling it data because it happens to contain a database would bury the thing the customer is actually paying for.

A cluster matching none of these (only Skip-Mapping types — an orphan Log Analytics workspace, a lone diagnostic setting) gets tier: "network_identity_secrets", since that is where observability lives and it needs no cutover slot of its own.

What the tier is used for

• Design orders clusters by tier (design.md § Step: Run the phase), so the network and data a compute cluster depends on are already mapped when the compute decision is made and the rubric’s Cluster Context criterion has something to read.
• Generate sequences the cutover by tier, which is the real payoff: it is the difference between a migration guide that says “create these 40 resources” and one that says what has to exist before what.
• The report groups by tier, so a reader sees the shape of the estate before the detail.

What the tier is NOT

Not a priority, not a difficulty estimate, and not a migration wave. A wave is a customer decision about risk and downtime that depends on business context this skill does not have. Tier is a dependency ordering; presenting it as a schedule would imply a plan nobody agreed to.

Status — build step 4

Implemented and consumed by discover-assemble.md. Generate’s use of it lands in step 6.