Case study: Il-Pastizzeria ta' Mario

This case study records an April 2026 run. Resource provisioning succeeded, but the recorded application checks returned HTTP 503. The generated records retain their original findings and are not a current deployment tutorial. Use the current workflow for agent handoffs and approvals.

Case study

Historical generated records

This case study preserves the April 14-15, 2026 artifacts. Editorial introductions and navigation have been refreshed; generated records remain verbatim. Resource provisioning succeeded, but production and staging checks returned HTTP 503.

IaC track: Bicep

This run used Bicep. It is not a verified Terraform exercise. Use the current workflow for either track’s main agents and gates. Historical automatic SKU substitution and review behavior are not instructions for a new deployment.

April 2026 case study

Read the decisions and results from a catering workload

Il-Pastizzeria ta' Mario recorded requirements, architecture, governance, planning, code generation, deployment, and as-built documentation during an April 2026 run.

Use this record to compare requirements, design decisions, code, and observed results. It does not describe every rule in current APEX.

Resource provisioning succeeded. Production and staging application checks returned HTTP 503. Read those limitations alongside the deployment summary rather than treating provisioning as application success.

Start the walkthrough Read the review findings

Workflow steps 7 + governance

Recorded reviews 4

IaC track Bicep + AVM

Primary region Sweden Central

Scenario

A Maltese catering outlet needs an online ordering app on Azure

The owner of a small catering outlet in Malta wants to sell pastizzi, Cisk, and Kinnie online. Payment is strictly cash on delivery. The app is a containerized React SPA backed by Azure Table Storage.

What the record contains

• Requirements and the original S1 design assumptions.
• Policy discovery and review findings from the recorded run.
• The P0v3 deployment result and unresolved application-health failures.

Walkthrough

Read the recorded steps

Open a step to inspect its generated records. Summaries below describe this case study, not the current workflow contract.

Step 1 Requirements Agent

Requirements

The requirements record describes online ordering for pastizzi, Cisk, and Kinnie, GDPR requirements, and a EUR 100-500 monthly budget.

Output

Business and technical requirements

Focus

Read the workload goals, compliance scope, and budget constraints used in the design.

Includes

• Business context
• Functional requirements
• Compliance & security
• Budget & scaling

Open step

Step 2 Architect Agent

Solution Architecture

App Service S1 with VNet integration and private endpoints. WAF scores: Security 8, Reliability 7, Performance 9, Cost 7, Operations 7. Estimated ~$155/month.

Output

WAF assessment and SKU recommendations

Focus

See the trade-offs between security posture, cost, and operational complexity before implementation begins.

Includes

• Service topology
• WAF assessment
• SKU recommendations

Open step

Step 3 Design Agent

Design Artifacts

Recorded ADRs cover App Service S1, Table Storage, and the revised private-endpoint design. The design estimate is distinct from the later as-built baseline.

Output

ADRs, diagrams, and cost visuals

Focus

Compare the recorded decisions, diagrams, and design-stage cost assumptions.

Includes

• Architecture diagram
• Architecture decisions
• Cost estimates

Open step

Step 3.5 Governance Agent

Governance

The April 2026 discovery recorded 21 policy assignments and a 9-tag resource-group blocker. Its audit-only findings describe that scope at that time.

Output

Azure Policy constraints

Focus

Compare the recorded policy scope, required tags, and unresolved constraints.

Includes

• Policy effects
• Required tags
• Deployment blockers

Open step

Step 4 IaC Planner

Infra-as-Code Plan

12 resources across 10 AVM Bicep modules deployed in 5 phases. Governance-adapted tag contract expands from 4 to 9 required tags.

Output

Implementation plan and dependency flow

Focus

Read the recorded module structure and phase dependencies.

Includes

• Module structure
• Implementation tasks
• Dependency flow

Open step

Step 5 Bicep CodeGen Agent

Infra-as-Code Gen

The CodeGen record lists 10 modules and reports build, lint, and source security checks. These are historical results, not a fresh validation.

Output

Bicep templates and validation results

Focus

Compare generated files with the plan and reported check results.

Includes

• File structure
• Validation results
• AVM modules

Open step

Step 6 Deploy Agent

Deployment

The April 2026 run provisioned 12 resources with azd and replaced unavailable S1 capacity with P0v3. Application checks returned HTTP 503. Current APEX requires approval for such changes.

Output

Deployment execution summary

Focus

Compare provisioning results, application-health failures, and the recorded SKU change.

Includes

• Preflight checks
• Deployment phases
• Outputs

Open step

Step 7 As-Built Agent

As-Built Docs

Post-deployment documentation: design document, resource inventory, operations runbook, backup/DR plan, compliance matrix, and cost estimate.

Output

Operational documentation suite

Focus

Read the observed inventory, cost assumptions, failed health checks, and open operational work.

Includes

• Design
• Resources
• Operations
• Compliance
• Cost

Open step

Reviews Challenger Agent

Adversarial Reviews

Four review records report 1 critical ACR SKU mismatch, 2 high, 10 medium, and 13 low findings.

Output

Cross-step findings

Focus

Inspect findings, severities, and recorded resolutions without assuming every issue was closed.

Includes

• Requirements
• Architecture
• Governance
• Implementation

Open step

Editorial scope

Historical outputs remain unchanged

Introductions, metadata, and navigation distinguish the historical run from current guidance. Generated artifact bodies retain their original wording.

What changed

The case-study framing states its date, observed failures, and differences from current approval rules.

What stayed intact

Generated records, cost figures, review findings, and imported assets. Their preservation is not a claim that every historical statement is correct.