Azure Blob Storage

Store evaluation results in Azure Blob Storage to enable team collaboration, centralized result browsing, and easy run comparison.

Why Azure Storage?

Without centralized storage:

• Results live on developers’ machines
• Hard to compare runs across team members
• No historical tracking
• Results deleted when workspace cleaned up

With Azure Storage:

• Automatic uploads — Results upload immediately after waza run
• Team visibility — Browse all runs with waza results list
• Easy comparison — Compare any two runs with waza results compare
• Historical archive — Results organized by date
• No extra config — Uses your existing Azure credentials

Prerequisites

Azure Account

You need an Azure Storage account. You can:

Option 1: Create a new storage account (recommended for team evaluation results)

# Create resource group
az group create -n waza-eval -l eastus

# Create storage account
az storage account create \
  -n myteamwaza \  # Must be unique globally, lowercase alphanumeric
  -g waza-eval \
  --sku Standard_LRS

# Create container
az storage container create -n waza-results -a myteamwaza

Option 2: Use existing storage account

# List your storage accounts
az storage account list --query "[].name"

# Use the account name in your .waza.yaml

Azure Authentication

Waza uses DefaultAzureCredential to automatically detect your Azure identity. It tries credentials in this order:

1. Environment variables (AZURE_CLIENT_ID, AZURE_CLIENT_SECRET, AZURE_TENANT_ID)
2. Managed Identity (when running on Azure VMs, App Service, etc.)
3. Azure CLI (az login)
4. Visual Studio Code (if you’ve signed in)
5. Azure PowerShell (if you’ve signed in)

In 99% of cases, just run:

az login

If you don’t have az CLI installed:

• macOS: brew install azure-cli
• Linux: curl -sL https://aka.ms/InstallAzureCLIDeb | sudo bash
• Windows: Download from Azure CLI releases

Configuration

Add a storage: section to your .waza.yaml file in your project root:

storage:
  provider: azure-blob
  accountName: "myteamwaza"    # Replace with your storage account name
  containerName: "waza-results" # Optional (default: waza-results)
  enabled: true                 # Optional (default: true when configured)

Configuration fields:

Field	Required	Description
provider	Yes	Must be azure-blob
accountName	Yes	Your Azure Storage account name (e.g., myteamwaza)
containerName	No	Blob container name. Default: waza-results
enabled	No	Toggle uploads on/off. Default: true when storage is configured

Getting Started

Step 1: Authenticate with Azure

az login

You’ll be guided to a browser to sign in with your Azure account.

Verify you’re logged in:

az account show

Step 2: Create a Storage Account (if needed)

If you don’t have a storage account yet:

# Create resource group
az group create -n waza-eval -l eastus

# Create storage account (account name must be globally unique, 3-24 chars, lowercase)
az storage account create \
  -n myteamwaza \
  -g waza-eval \
  --sku Standard_LRS

# Create container
az storage container create -n waza-results -a myteamwaza

Note the storage account name — you’ll use it in .waza.yaml.

Step 3: Configure .waza.yaml

cat > .waza.yaml <<EOF
storage:
  provider: azure-blob
  accountName: "myteamwaza"
  containerName: "waza-results"
  enabled: true
EOF

Replace myteamwaza with your actual storage account name.

Step 4: Run Evaluations

waza run evals/my-skill/eval.yaml -v

During the run, you’ll see:

📤 Uploading results to Azure Blob Storage...
✅ Results uploaded: waza-results/code-explainer/20250226-143025-abc123.json

Using Results from Storage

List All Runs

Browse all uploaded evaluation results:

waza results list

Output:

ID                          Skill           Model              Status   Timestamp
20250226-143025-abc123     code-explainer  claude-sonnet-4.6  passed   2025-02-26 14:30:25
20250226-135410-def456     code-explainer  gpt-4o             passed   2025-02-26 13:54:10
20250226-124800-ghi789     math-solver     claude-sonnet-4.6  failed   2025-02-26 12:48:00

Show more results:

waza results list --limit 50

Export as JSON:

waza results list --format json | jq .

Compare Two Runs

Compare results side-by-side to see what changed:

waza results compare 20250226-143025-abc123 20250226-135410-def456

Output shows:

• Task-level score changes
• Pass/fail differences
• Key metrics (tokens used, time taken, etc.)

Example:

Task Comparison
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Task                      Score Δ    Status
basic-explanation         -5 (-2%)   ↓ regressed
edge-case-handling        +10 (+5%)  ↑ improved
performance-check         0 (0%)     ↔ unchanged

Aggregate Metrics
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Pass rate:                85% → 90% (+5%)
Avg completion time:      1.2s → 0.9s (-25%)
Tokens used:              2,450 → 2,380 (-3%)

View comparison as JSON:

waza results compare run-id-1 run-id-2 --format json

Full Example Workflow

Here’s a complete example using Azure Storage for team evaluation:

# 1. Set up Azure (one-time)
az login
az group create -n waza-eval -l eastus
az storage account create -n myteamwaza -g waza-eval --sku Standard_LRS
az storage container create -n waza-results -a myteamwaza

# 2. Configure project (one-time)
cat > .waza.yaml <<EOF
storage:
  provider: azure-blob
  accountName: "myteamwaza"
  containerName: "waza-results"
  enabled: true
EOF

# 3. Run evaluations (repeatable)
waza run evals/code-explainer/eval.yaml -v -o results-$(date +%s).json

# 4. Browse team's results
waza results list --limit 20

# 5. Compare your run with previous run
waza results compare 20250226-143025-abc123 20250226-135410-def456

Troubleshooting

❌ Error: “Subscription not found”

Cause: az login didn’t work or you don’t have Azure access.

Fix:

# Clear cached credentials
az logout

# Re-authenticate
az login

# Verify access
az account show

❌ Error: “Storage account not found”

Cause: Account name in .waza.yaml doesn’t exist or has a typo.

Fix:

# List your storage accounts
az storage account list --query "[].name"

# Update .waza.yaml with the correct name

❌ Error: “Container not found”

Cause: The blob container doesn’t exist in the storage account.

Fix:

# Create the container (replace account name)
az storage container create -n waza-results -a myteamwaza

❌ Error: “Access Denied”

Cause: Your Azure user doesn’t have storage permissions.

Fix: Contact your Azure admin to grant you the Storage Blob Data Contributor role on the storage account.

❌ Results not uploading

Fix: Verify storage is enabled and configured:

cat .waza.yaml | grep -A 5 "^storage:"

Upload is automatic only if:

• storage: section exists
• enabled: true (or omitted, since it defaults to true)
• You’ve run az login

❌ Where did my results go?

Results are stored in your Azure Storage account:

# View results in storage
az storage blob list -c waza-results -a myteamwaza

# Download a specific result
az storage blob download \
  -c waza-results \
  -n code-explainer/20250226-143025-abc123.json \
  -a myteamwaza \
  -f result.json

Advanced

Disable uploads for a single run

# Create a temporary config with storage disabled
cat > .waza-local.yaml <<EOF
storage:
  enabled: false
EOF

# Run with local-only results (doesn't upload)
waza run eval.yaml -v

Use a custom Azure credential

For CI/CD or service accounts, set environment variables:

export AZURE_CLIENT_ID="your-app-id"
export AZURE_CLIENT_SECRET="your-secret"
export AZURE_TENANT_ID="your-tenant-id"

waza run eval.yaml  # Uses these credentials

Next Steps

• CLI Reference — Complete waza results command docs
• YAML Schema — Full storage configuration reference
• Dashboard Guide — Visualize results in the web dashboard