Team Split Assessment

Plan Under Review

Source: Development Blueprint (Steps 1–5)
Date: 2026-03-18

Complexity Analysis

Factor	Value	Threshold	Assessment
Total tasks	~25	15	Over
Distinct phases with sequential dependencies	4 (infra → workflows → validation → cutover)	2	Over
Distinct persona types needed	3 (CDK/CFn engineer, GHA workflow author, operator)	4	Under
Estimated working directories	3 repos (`infrastructure`, `arda-frontend-app`, `documentation`)	1	Over
Cross-cutting concerns (handoffs, gates)	3 (IAM role → workflow OIDC, CFn exports → workflow reads, validation → cutover GO/NO-GO)	1	Over
Tasks with external validation (builds, tests, deploys)	8+ (CDK synth, CFn deploy, OIDC test, workflow trigger, Amplify build, site verification, PR preview)	5	Over

Qualitative Factors

Context window pressure: High — tracking CloudFormation templates, CDK constructs, bash scripts, GitHub Actions YAML, and AWS CLI verification across 3 repos.
Error blast radius: High — a misconfigured IAM role invalidates all workflow development; a bad cutover step can break production.
Natural phase boundaries: Clear checkpoints — demo app deployed, role assumable, workflow working, validation passed, cutover complete.

Recommendation

Split into 4 runs. The project has clear sequential dependencies (infrastructure must exist before workflows can use it), spans 3 repositories, and has a critical GO/NO-GO checkpoint between validation and production cutover. Splitting isolates blast radius and provides natural verification points.

Proposed Runs

Run	Directory	Description	Tasks	Repos	Est. Duration
1	`run-1-infrastructure/`	Demo environment setup + IAM role creation	8	`infrastructure`, `arda-frontend-app` (branch only)	1 session + deployment
2	`run-2-workflows/`	GitHub Actions workflow development on `demo` branch	5	`arda-frontend-app`	1 session
3	`run-3-validation/`	End-to-end pipeline validation on demo partition	5	`arda-frontend-app` (triggers only)	1 session + manual verification
4	`run-4-cutover/`	Production migration of dev, stage, prod + PR previews + documentation	12	`infrastructure`, `arda-frontend-app`, `documentation`	2 sessions + staged rollout

Artifact Flow

Run 1 (Infrastructure)
  Produces:
    - demo branch in arda-frontend-app
    - Alpha001-demo-Amplify + Alpha001-demo-AmplifyBranch CloudFormation stacks
    - CloudFormation exports: Alpha001-demo-I-AmplifyAppId, Alpha001-demo-I-AmplifyBranchName
    - IAM role: Alpha001-API-GitHubActionFrontEnd, Alpha002-API-GitHubActionFrontEnd
    - amm.sh with mappings, gate, and export stack support
        ↓
Run 2 (Workflows)
  Consumes:
    - demo branch (to push workflow files)
    - IAM role (for OIDC assumption)
    - CloudFormation exports (for Amplify parameters)
  Produces:
    - deploy.yaml, redeploy.yaml, reusable_deployment.yaml on demo branch
    - GitHub environments configured
        ↓
Run 3 (Validation)
  Consumes:
    - Working workflows on demo branch
    - Functional demo Amplify app
  Produces:
    - Validated pipeline (deploy + redeploy confirmed working)
    - GO/NO-GO decision for cutover
        ↓
Run 4 (Cutover)
  Consumes:
    - Validated workflows (merged to main)
    - amm.sh with export stack support
  Produces:
    - Lightweight export stacks for dev, stage, prod
    - All partitions migrated to new pipeline
    - PR preview deployments enabled on dev app
    - Frontend development guide
    - Post-cutover instructions

Risk Analysis

Risk	Mitigation via Split
IAM role misconfiguration blocks all workflow development	Run 1 validates OIDC assumption before Run 2 starts
Workflow bug breaks demo deployment	Run 2 is isolated to demo; existing pipeline unaffected
Cutover breaks production	Run 3 provides explicit GO/NO-GO gate before Run 4
Region mismatch for prod (`us-east-2` vs `us-east-1`)	Run 4 handles prod last; dev/stage verified first
Kyle regression from infrastructure changes	Run 1 exit criteria include Kyle regression check

Open Questions and Decisions

#	Question	Options	Recommendation	Decision
S1	Should Run 1 deploy to both accounts upfront or defer Alpha002?	(a) Both upfront, (b) Alpha001 only, Alpha002 at cutover	(a) Both upfront — per design decision C2, the role is part of infrastructure	(a)
S2	Should Run 3 validation include automated E2E tests or manual verification only?	(a) Manual only, (b) Playwright MCP, (c) Existing E2E suite	(a) Manual — user signs in, agent verifies page navigation	Agreed
S3	Should Run 4 migrate partitions in a single session or across multiple sessions with pauses?	(a) Single session, (b) One partition per session	(b) One partition per session for safety	Agreed