Test Plan for 3D Engagement Cube

Coverage Target: 70% code coverage Testing Framework: pytest 7.4.0 Python Version: 3.13 OS: macOS 14.6

Part 1: Ideal Test Plan (Unlimited Time)

Unit Testing (90% coverage target)

Data Layer (src/data_loader.py)

• Mock S3 connections and error handling

• CSV parsing with malformed data

• Date normalization across formats

• Account ID standardization

Feature Engineering (src/feature_build.py)

• 90-day Interest window validation

• 270-day Intent window validation

• Aggregation with missing/null values

• Edge case handling (empty accounts)

Model Training (src/train.py)

• Model reproducibility (same seed → same output)

• Calibration procedures

• Model serialization/deserialization

Scoring (src/score.py, src/intent_scorer.py, src/icp_scorer.py)

• Tier assignments (Low/Medium/High at 33/66 thresholds)

• Score normalization (0-100 scale)

• Overall score weights (40/40/20)

• SHAP explanation generation

Integration Testing

• Full pipeline: S3 → features → training → scoring → hot-list

• Test with 100, 1K, 10K account datasets

• Config parameter propagation

• Output format validation (CSV, Parquet)

System Testing

• Performance benchmarking (training time, memory usage, SHAP performance)

• Reliability (network failures, corrupted data, out-of-memory)

• Historical validation against actual conversions

• Class imbalance handling

Acceptance Testing

• Sales team validates top 20 accounts

• SHAP explanations interpretability

• A/B test against current method

• ROI measurement

Part 2: Actual Test Plan

Unit Tests

Test Files:

tests/

├── test_data_loader.py

├── test_feature_build.py

├── test_train.py

├── test_score.py

├── test_icp_scorer.py

└── test_intent_scorer.py

Priority Test Cases:

test_data_loader.py:

• Account ID normalization (string/float handling)

• Date parsing (multiple formats)

• Missing column handling

test_feature_build.py:

• 90-day Interest window calculation

• 270-day Intent window calculation

• Empty account handling

test_train.py:

• Model reproducibility

• Score range (0-100)

test_score.py:

• Tier thresholds (33/66)

• Overall score calculation (40/40/20)

test_intent_scorer.py:

• Rule-based scoring logic

• Reason generation

test_icp_scorer.py:

• ICP tier assignments

• Firmographic scoring rules

Run Tests:

make test  # Runs: pytest tests/ -v --tb=short

Environment: Python 3.13, pytest 7.4.0, macOS 14.6

Integration Testing

Full Pipeline Test:

make full-pipeline

Test Scenarios:

1. Small dataset (100 accounts)

2. Edge cases (no activity, missing data)

3. S3 vs local data source

Validation:

• Pipeline completes without errors

• Outputs generated in artifacts/scores/

• Hot list contains top accounts

Environment: Same as unit tests + AWS S3 credentials

Error Conditions

Scenarios:

• Missing S3 data (fallback to local)

• Duplicate account IDs

• Invalid date formats

• High % missing values

Expected: Graceful error messages, clear logging

Test Commands

# Unit tests

make test

# Integration test

make full-pipeline

# Calibration validation

make calibrate

# Pipeline status

make status

Success Criteria

1. 70%+ unit test coverage, all passing

2. Full pipeline runs without errors

3. Model calibration passes validation

4. Performance < 30 min end-to-end

5. Sales team validates top accounts

Dependencies

Required:

• Python 3.13

• pytest 7.4.0

• pandas 2.0.0+

• scikit-learn 1.3.0+

• lightgbm 4.0.0+

• shap 0.42.0+

• boto3 1.28.0+ (for S3)

Install: make install