Validation Rules Reference

This page defines the complete set of validation rules for a CML AMP project conforming to the Distributed XGBoost with Dask pattern. Rules are categorized by domain and severity.

Severity levels:

ERROR — the artifact is invalid and will fail at runtime.
WARNING — the artifact may work but deviates from the expected contract.

Structural Rules

Rule	Severity	Description
`S-001`	ERROR	Repository must contain `.project-metadata.yaml` at root
`S-002`	ERROR	`.project-metadata.yaml` must parse as valid YAML
`S-003`	ERROR	`requirements.txt` must exist at root
`S-004`	ERROR	`scripts/predict_fraud.py` must exist
`S-005`	ERROR	`model/best-xgboost-model` must exist (post-training only)
`S-006`	ERROR	`cdsw-build.sh` must exist
`S-007`	ERROR	`utils/dask_utils.py` must exist
`S-008`	ERROR	`setup.py` must exist at root

Dependency Rules

Rule	Severity	Description
`D-001`	ERROR	`requirements.txt` must include `xgboost`
`D-002`	ERROR	`requirements.txt` must include `dask` (with or without `[complete]` extra)
`D-003`	ERROR	`requirements.txt` must include `scikit-learn`
`D-004`	ERROR	`requirements.txt` must include `-e .` (local utils package)
`D-005`	ERROR	`requirements.txt` must include `numpy`
`D-W01`	WARNING	All packages should have pinned versions (`==`) for reproducibility

Model Endpoint Rules

Rule	Severity	Description
`E-001`	ERROR	`scripts/predict_fraud.py` must parse as valid Python (no syntax errors)
`E-002`	ERROR	`scripts/predict_fraud.py` must define a function named `predict_fraud`
`E-003`	ERROR	`predict_fraud` must accept exactly one parameter (`args`)
`E-004`	ERROR	Module must load a model from `/home/cdsw/model/best-xgboost-model`
`E-005`	ERROR	Module must define a `threshold` variable
`E-W01`	WARNING	`threshold` should be a float between 0.0 and 1.0
`E-W02`	WARNING	`predict_fraud` should return an integer (0 or 1)

AMP Configuration Rules

Rule	Severity	Description
`A-001`	ERROR	`.project-metadata.yaml` must contain a `runtimes` list with at least one entry
`A-002`	ERROR	Runtime `kernel` must specify Python 3.9 or higher
`A-003`	ERROR	Runtime `editor` must be `JupyterLab`
`A-004`	ERROR	At least one task must reference `scripts/install_dependencies.py`
`A-W01`	WARNING	`specification_version` should be `1.0`

Cluster Utility Rules

Rule	Severity	Description
`C-W01`	WARNING	`utils/dask_utils.py` should define `run_dask_cluster`
`C-W02`	WARNING	`run_dask_cluster` should accept `num_workers`, `cpu`, `memory` parameters
`C-W03`	WARNING	`run_dask_cluster` should return a dict with keys: `scheduler`, `workers`, `scheduler_address`, `dashboard_address`
`C-W04`	WARNING	Scheduler should listen on TCP port 8786

Build Script Rules

Rule	Severity	Description
`B-001`	ERROR	`cdsw-build.sh` must be a valid shell script (starts with a command, not a syntax error)
`B-W01`	WARNING	`cdsw-build.sh` should install dependencies from `requirements.txt`