Eighth phase of Arc 3. Integration, not new content. Multi-cloud basecamp. The repo goes public.

Phases 17-23 added a lot: kernel internals, networking, containers from scratch, K8s, distributed systems, IaC, AWS. This phase doesn’t add an eighth thing. It proves the previous seven transferred. GCP is the compare-cloud step at year scale. Multi-cloud basecamp is the integration. basecamp going public is the artifact.

This is also the moment Arc 3 stops being internal practice and starts being a portfolio. basecamp flips from private to public. Anyone landing on its GitHub repo now sees a multi-cloud GitOps platform with forge + beacon + observability + mesh. That’s the senior-engineer signal.

Prerequisites

• All of Phases 17-23 complete
• GCP $300 credits available
• basecamp substrate + forge alive on homelab + AWS
• You accept: this phase is the proof, not the lesson. Lessons happened in Phases 17-23. Here you make them visible.

Why this phase exists

A capstone integration phase serves two purposes. First, it forces previous phases’ work to actually compose: most engineering pain happens at integration boundaries. Second, it produces a visible artifact (the public basecamp repo) that proves Arc 3 happened.

The GCP compare is the year-scale instance of the COMPARE step. If GCP feels easy after Phase 23’s AWS, the pattern transferred. If GCP feels foreign, AWS taught you AWS-specific naming.

The pattern-first frame

Same eight steps.

1. PROBLEM

You have a working platform on homelab + AWS. You have IaC + GitOps + paved-road discipline. The question: did the patterns transfer to a different cloud? If yes, basecamp runs on GCP with minimal changes. If no, you’ll discover AWS-specific assumptions leaked into the platform.

2. PRINCIPLES

2.1 Multi-tenancy

A multi-tenant platform serves multiple isolated workloads on shared substrate. At Arc 3 scale, “multi-cloud” is a form of multi-tenancy.

→ Pattern: multi-tenancy: first OUTLINE

Investigate:

• What does isolation mean across clouds (identity, network, data, control plane)?
• When does multi-cloud add real resilience vs complexity tax?
• What changes about developer experience with 3 clouds vs 1?

2.2 Cloud abstraction (the limits)

You discovered in Phase 22 that IaC doesn’t actually abstract clouds. The platform layer has more abstraction potential: the developer says “deploy this service” and the platform decides which cloud.

Investigate:

• What does ascent new service need to do differently for GKE vs EKS?
• Which basecamp services can run anywhere, and which are pinned?
• Where does AWS-specific assumption leak into the platform layer?

2.3 Cost across clouds

Each cloud has different pricing curves. Egress expensive on AWS, less on GCP. GPU instances easier to reserve on GCP for short bursts. Storage costs vary.

Investigate:

• Where does basecamp cost-optimize for AWS in ways that won’t transfer?
• What’s the right cost-distribution strategy for production on both?
• Why is cross-cloud egress almost always the largest unexpected cost?

2.4 Capstone discipline

A capstone is subtractive. What do you cut to keep the narrative clear?

Investigate:

• Smallest GCP services you need to prove the pattern transferred?
• What does basecamp’s README need to look like for a stranger to grok in 5 minutes?
• Which substrate + forge components don’t need to be visible in the public repo?

3. TRADE-OFFS

4. TOOLS (as of 2026-06)

• gcloud CLI
• GKE Standard (not Autopilot; operational visibility matters)
• Cloud SQL for Postgres
• GCS for object storage
• Secret Manager for secrets

Reading

• “Google Cloud Platform in Action” (Geewax)
• GCP docs: Foundations and Networking concepts

5. MASTERY: Multi-cloud basecamp

5.1 The integration

• Provision small GKE via your forge GCP modules
• Connect GKE to basecamp’s ArgoCD as a target
• Deploy beacon to GKE in parallel with EKS and K3s
• Wire Cilium Mesh across all three (or use federation pattern)
• Update ascent to support --target gke|eks|k3s (Phase 26 deepens ascent)
• Verify ascent new service hello --target gke works end-to-end

5.2 Going public

Flip basecamp’s repo from private to public. Before flipping:

• No secrets in Git history (git-secrets or trufflehog scan)
• Strong README: what is basecamp, what does it run, who is it for, how to read it
• LICENSE chosen (Apache 2.0 or MIT)
• Architecture diagram included
• ADRs published

5.3 Operational depth checklist

[ ] GKE small cluster provisioned via forge; destroyed cleanly
[ ] ArgoCD multi-cluster: same Application synced to both EKS and GKE
[ ] ascent --target gke working end-to-end (foreshadow Phase 26)
[ ] basecamp repo scanned for secrets; clean
[ ] basecamp public on GitHub; README explains in 5 minutes
[ ] Two services running on all 3 clouds; same Helm chart, different cluster targets
[ ] Cost dashboard showing AWS + GCP + homelab side-by-side
[ ] Demo: deploy a new service via ascent to all 3 clouds in one command

6. COMPARE: Cost analysis

Run the same workload on AWS, GCP, homelab for 1 week. Capture cost. Build a small markdown table showing total cost per service per cloud.

400-word reflection: where did surprises come from? Predictable from docs vs not? Single cloud for next employer, and which and why?

7. OPERATE

• 3-4 runbooks: multi-cloud
• 1-2 ADRs (EKS as primary, GKE as DR, or similar)
• Weekly log

8. CONTRIBUTE

• ArgoCD multi-cluster docs
• Cilium federation / cluster-mesh docs
• gcloud CLI corpus

What ships from this phase

• basecamp repo goes public: the most consequential ship of the year
• GKE-side of basecamp alive, parallel to EKS and K3s
• Architecture diagram + ADRs published in basecamp/docs
• Arc 3’s public artifact set: forge, beacon, basecamp

Validation criteria

[ ] Multi-cloud basecamp working (K3s + EKS + GKE)
[ ] basecamp repo public; README + LICENSE + architecture diagram
[ ] ascent --target flag working for all 3 clouds (foreshadow Phase 26)
[ ] All 8 operational depth checks
[ ] Cost comparison reflection (400 words)
[ ] 3-4 multi-cloud runbooks
[ ] Pattern entries:
    - multi-tenancy → first OUTLINE
    - all prior Arc 3 patterns reinforced toward DEEP
[ ] Exit Test passed

Exit Test

Time: 3 hours.

Part 1: Build (90 min)

Provision fresh GKE cluster via forge; connect to ArgoCD; deploy hello to it. Verify mTLS works between hello on GKE and beacon on EKS via mesh federation (or document why federation is too expensive at this scale).

Part 2: Articulate (90 min)

~1200 words: “basecamp is now multi-cloud. Defend the choice. Cover developer experience, operational cost, resilience, vendor lock-in, and what specifically you’d change to add a third cloud (Azure). Cite Arc 3 patterns.”

Anti-patterns

Patterns touched this phase

• multi-tenancy: first OUTLINE
• All Arc 3 patterns reinforced toward DEEP

→ Next: Phase 25: Service Mesh + Zero-Trust Networking