No hardcoded hyperparams

Seed set comprehensively

MLflow tracking used

Required params logged

Required metrics logged

Model artifact logged

Config artifact logged

MLflow tags used

Training/inference separation

training_report.json written

requirements.txt present

sklearn ColumnTransformer used

Numeric imputer present

StandardScaler for numeric

Categorical imputer present

OneHotEncoder for categorical

Data validation before transform

joblib serialization used

Metadata JSON saved alongside pipeline

Descriptive feature naming

pipeline_summary.json written

Pipeline artifact in artifacts/ dir

Retry config on tasks

Retry delay configured

Conditional deployment

Explicit accuracy threshold

Failure alerting configured

Idempotent stage design

Training/inference separation

No credentials in code

pipeline_design.md present

No silent failure

Execution timeout set

Model signature inferred

registered_model_name set

Staging stage used

Production promotion

Archive existing on promotion

No promotion without metrics

Validation threshold check

MLflow tags on runs

Model comparison documented

registry_report.json written

requirements.txt present

Optuna used

MedianPruner configured

Pruner startup trials = 5

Pruner warmup steps = 3

SQLite storage used

load_if_exists=True

trial.report called

trial.should_prune checked

Best params logged to MLflow

best_params.json written

requirements.txt present

StratifiedKFold used

CV mean and std reported

McNemar's test applied

p-value computed and reported

Explicit ValidationStatus or equivalent

Accuracy threshold defined as variable

No deployment without metrics

Segment-level evaluation present

validation_report.json written

requirements.txt present

Reproducibility: seeds set