Title: Lifecycle Management
Locale: en
URL: https://sensorswave.com/en/docs/experiments/lifecycle-management/
Description: Master the full experiment workflow management
Experiment lifecycle management covers the entire workflow from creation, debugging, running, monitoring, to completion. Scientific lifecycle management ensures experiments proceed smoothly and produce reliable conclusions. This article details the key management points for each stage.

## Experiment lifecycle

An experiment goes through the following 6 statuses during its lifecycle:

| Status | English | Description |
|--------|---------|-------------|
| Draft | DRAFT | Experiment has been created but not published; configuration can be modified repeatedly |
| Debugging | DEBUG | For test environment validation; visible only to specific test users |
| Running | RUNNING | Experiment is officially live, collecting data and performing splits |
| Released | RELEASED | Winning Variant is applied to all users |
| Finished | FINISHED | Experiment manually ended; no new data is collected |
| Deleted | DELETED | Experiment is marked for deletion |

### Status transitions

```mermaid
stateDiagram-v2
    [*] --> DRAFT
    DRAFT --> DEBUG : Enter debugging
    DEBUG --> RUNNING : Officially release
    RUNNING --> RELEASED : Release winning Variant to all users
    RUNNING --> FINISHED : Manually end
    RELEASED --> FINISHED : End experiment
    FINISHED --> DELETED : Archive and delete
    DRAFT --> DELETED : Delete directly

    DRAFT : Draft
    DEBUG : Debugging
    RUNNING : Running
    RELEASED : Released
    FINISHED : Finished
    DELETED : Deleted
```

---

## DRAFT stage

### Characteristics

- Experiment has been created but not released
- Configuration can be modified repeatedly
- Experiment configuration cannot be retrieved in code
- No exposure logs are recorded

### Operations

**Edit experiment**:
- Modify basic information (display name, description, Hypothesis)
- Adjust Variant configuration (Allocation, variable values)
- Modify Targeting Rules and experiment Metrics

**Delete experiment**:
- If the experiment configuration is incorrect, you can delete it directly
- Deletion changes the status to DELETED

**Enter debugging**:
- After confirming the configuration, switch the experiment to DEBUG status

---

## DEBUG stage

### Characteristics

- Visible only to specific test users
- Used to validate experiment configuration and code integration in the test environment
- Does not affect production traffic

### Operations

**Validate experiment configuration**:
- Use test accounts to trigger the experiment
- Confirm each Variant displays correctly
- Verify exposure logs are reported normally

**Official release**:
- After debugging validation passes, switch the experiment status to RUNNING

---

## RUNNING stage management

### Monitor traffic distribution

After release, first verify that the Allocation is even.

**How to view**:

Query the `$ABImpress` Event in Segmentation:

```
Event: $ABImpress
Filter: experiment_key = 'your_experiment_key'
Group by: variant
Metric: Unique users
Time range: Last 1 day
```

**Expected result** (50/50 Allocation):

| Variant | Users | Percentage |
|---------|-------|------------|
| control | ~500 | ~50% |
| treatment | ~500 | ~50% |

**Handling anomalies**:

If the distribution is severely uneven (deviation > 5%):
1. Check if the Allocation configuration is correct
2. Check if the Targeting Rules are too restrictive
3. Wait for more samples — short-term fluctuations are normal

### Monitor exposure logs

Ensure exposure logs are being reported normally.

**Verification steps**:

1. View exposure data for the last 1 hour
2. Confirm each user has only one exposure log
3. Check that exposure properties are complete

**Handling anomalies**:

If exposure logs are not being recorded:
- Check if the SDK has A/B testing enabled
- Check if the experiment has been released
- Check if code integration is correct
- Check if the user meets the Targeting Rules

### Observe key Metrics

Regularly check the trend of core Metrics.

**Primary Metric**:

Check the primary Metric daily:

```
Event: Conversion event (for example, AddToCartClicked)
Filter: experiment = 'your_experiment_key'
Group by: variant
Metric: Conversion rate
Time range: Cumulative
```

**Secondary and guardrail Metrics**:

Check every 2–3 days:
- Ensure secondary Metrics are normal
- Ensure guardrail Metrics have no negative impact

### Handling anomalies

**When anomalous Metrics are detected**:

If any of the following occur, end the experiment immediately:
- Severe decline in guardrail Metrics (for example, error rate > 5%)
- Severe decline in primary Metrics (for example, conversion rate drops > 20%)
- Surge in user complaints

---

## Experiment decision

### When to end an experiment

End the experiment when any of these conditions are met:

**Success conditions**:
- ✅ Reached predetermined sample size (> 1,000 users per group)
- ✅ Test Group primary Metric significantly outperforms Control
- ✅ Statistically significant: p  1,000 users per group
- Run for at least 1 week (covering weekdays and weekends)
- Reach statistical significance (p < 0.05)

### Q: Can an experiment end early?

**A**: Not recommended. Ending too early may result in:
- Insufficient sample size, unreliable results
- Short-term fluctuations mistaken for long-term effects

**Exceptions**:
- Severe bug discovered
- Guardrail Metrics severely decline

### Q: How long should experiment code be kept after the experiment ends?

**A**:
- Experiment logic: Remove immediately
- Winning solution code: Keep long-term
- Experiment tracking code: Can be removed (keep regular tracking)

### Q: What is the difference between RELEASED and FINISHED?

**A**:
- **RELEASED**: The selected winning Variant is applied to all users; all users will see that Variant
- **FINISHED**: The experiment is manually ended; no new data is collected; no Variant has been applied to all users

---

## Related documentation

- [Create and Configure](create-and-configure.mdx): Learn how to create experiments
- [Metrics and Analysis](metrics-and-analysis.mdx): Learn how to analyze experiment results
- [Best Practices](best-practices.mdx): Master experiment management best practices

---

**Last updated**: January 29, 2026