Time-Shifted Execution

Defer non-urgent requests to execute during periods of lower grid carbon intensity.

How Time-Shifting Works

Time-shifted execution analyzes grid carbon intensity forecasts and schedules your requests for optimal sustainability:

1. Submit Request: Queue your non-urgent workload
2. Carbon Forecast: System predicts grid carbon intensity
3. Optimal Scheduling: Request executes when carbon intensity is lowest
4. Result Delivery: Receive response when complete

Basic Usage

from openai import OpenAI

client = OpenAI(
    base_url="https://api.gateflow.ai/v1",
    api_key="gw_prod_your_key_here"
)

# Defer execution to low-carbon period
response = client.chat.completions.create(
    model="gpt-5.2",
    messages=[{"role": "user", "content": "Generate comprehensive sustainability report"}],
    routing_mode="time_shift_optimized",
    max_delay_hours=12,  # Maximum acceptable delay
    priority="normal"  # low, normal, high
)

print(f"Job ID: {response.deferred_job_id}")
print(f"Estimated completion: {response.estimated_completion}")
print(f"Expected carbon savings: {response.estimated_carbon_savings_gco2e} gCO₂e")

Configuration Options

Delay Settings

# Short delay for slightly better carbon conditions
response = client.chat.completions.create(
    model="auto",
    messages=[{"role": "user", "content": "Quick analysis"}],
    routing_mode="time_shift_optimized",
    max_delay_hours=2,  # Complete within 2 hours
    min_carbon_savings_gco2e=10  # Minimum savings to justify delay
)

# Long delay for maximum carbon savings
response = client.chat.completions.create(
    model="auto",
    messages=[{"role": "user", "content": "Overnight data processing"}],
    routing_mode="time_shift_optimized",
    max_delay_hours=24,  # Can wait up to 24 hours
    target_carbon_intensity=200  # Wait for grid intensity <200 gCO₂/kWh
)

Priority Levels

Priority	Typical Delay	Use Case
low	6-24 hours	Background processing, batch jobs
normal	2-12 hours	Reports, non-urgent analysis
high	0-4 hours	Time-sensitive but flexible tasks

Monitoring Deferred Jobs

# Check job status
job_status = client.deferred_jobs.get(response.deferred_job_id)

print(f"Status: {job_status.status}")
print(f"Scheduled for: {job_status.scheduled_for}")
print(f"Carbon intensity at submission: {job_status.grid_intensity_at_submission}")
print(f"Carbon intensity at execution: {job_status.grid_intensity_at_execution}")
print(f"Actual carbon savings: {job_status.carbon_saved_vs_immediate}")

# Wait for completion and get result
if job_status.status == "completed":
    result = client.deferred_jobs.result(response.deferred_job_id)
    print(f"Result: {result.content}")

Batch Processing with Time-Shifting

# Process multiple requests in optimized batch
batch_response = client.batch.create(
    requests=[
        {
            "model": "auto",
            "messages": [{"role": "user", "content": "Analyze dataset A"}]
        },
        {
            "model": "auto",
            "messages": [{"role": "user", "content": "Analyze dataset B"}]
        },
        {
            "model": "auto",
            "messages": [{"role": "user", "content": "Analyze dataset C"}]
        }
    ],
    routing_mode="time_shift_optimized",
    max_delay_hours=8,
    batch_optimization="carbon_efficient"  # Optimize batch execution order
)

print(f"Batch ID: {batch_response.batch_id}")
print(f"Estimated total carbon savings: {batch_response.estimated_total_savings_gco2e}")

Carbon Savings Potential

Typical Savings by Delay

Delay Period	Typical Carbon Savings	Best For
2-4 hours	20-40%	Short-term optimization
4-8 hours	40-60%	Daily workload shifting
8-12 hours	60-75%	Overnight processing
12-24 hours	75-85%	Weekend/off-peak execution

Regional Variations

Carbon savings vary by region based on grid mix:

• Nordics/Hydro-rich: 30-50% savings
• France/Nuclear: 40-60% savings
• US Mixed Grid: 50-70% savings
• Coal-heavy Regions: 70-85% savings

Best Practices

Ideal Use Cases

✅ Perfect for:

• Nightly data processing
• Report generation
• Batch analytics
• Content creation pipelines
• Non-urgent customer communications
• Background data enrichment

❌ Avoid for:

• Real-time customer interactions
• Time-critical alerts
• Financial transactions
• Emergency response systems
• User-facing real-time features

Implementation Strategy

1. Start Small: Test with non-critical workloads
2. Monitor Impact: Track carbon savings and completion times
3. Adjust Thresholds: Find optimal delay/savings balance
4. Expand Gradually: Apply to more workloads as confidence grows
5. Combine Approaches: Use with Sustain Mode for maximum impact

Advanced Configuration

Carbon Intensity Targets

# Set specific carbon intensity thresholds
response = client.chat.completions.create(
    model="auto",
    messages=[{"role": "user", "content": "Process when grid is cleanest"}],
    routing_mode="time_shift_optimized",
    max_delay_hours=12,
    target_carbon_intensity=150,  # Wait for <150 gCO₂/kWh
    min_savings_percentage=50  # Require at least 50% savings vs immediate execution
)

Regional Preferences

# Prioritize specific low-carbon regions
response = client.chat.completions.create(
    model="auto",
    messages=[{"role": "user", "content": "Execute in greenest region"}],
    routing_mode="time_shift_optimized",
    max_delay_hours=6,
    preferred_regions=["nordics", "canada", "france"],
    avoid_regions=["coal-heavy"]
)

Troubleshooting

"Job not executing within expected time"

Causes:

• Carbon intensity not dropping as predicted
• Grid conditions more volatile than usual
• Regional constraints limiting options

Solutions:

1. Increase max_delay_hours
2. Lower target_carbon_intensity
3. Reduce min_savings_percentage
4. Check regional grid conditions

"Carbon savings lower than estimated"

Causes:

• Grid conditions changed after estimation
• Regional carbon intensity higher than expected
• Job executed before optimal conditions

Solutions:

1. Increase maximum delay for better optimization
2. Use more flexible carbon intensity targets
3. Monitor grid conditions in your dashboard

Time-Shifting vs Other Optimization Methods

Method	Carbon Savings	Latency Impact	Best For
Sustain Mode	30-50%	Minimal	Real-time requests
Time-Shifting	50-85%	Significant	Non-urgent workloads
Caching	80-90%	None	Repeated requests
Combined	70-95%	Varies	Comprehensive optimization

Next Steps

• Explore Sustain Mode - Real-time carbon optimization
• View Deferred Jobs - Monitor time-shifted requests
• Configure Regional Policies - Optimize for specific regions
• Set Up Alerts - Get notified of optimal execution windows