Code Convention

General Principles

• Use Panel for concrete data.
• Use Graph for lazy derived logic.
• Put domain operations in transformer and composer modules.
• Prefer function-style APIs.
• Keep operations deterministic and stateless.
• Do not mutate input frames in place.
• Keep Panel immutable at the public boundary.

Public API Rule

Transformer: Panel | Graph -> Graph
Composer:    (Panel | Graph, ...) -> Graph
Execution:   Graph -> Panel output

Do not add one Graph method per operation.

Custom Operations

Use decorators:

@transformer
def rolling_mean(frame: pd.DataFrame, *, window: int) -> pd.DataFrame:
    return frame.rolling(window).mean()


@composer
def average(*frames: pd.DataFrame) -> pd.DataFrame:
    return sum(frames) / len(frames)

Operation configuration must be deterministic and serializable.

Internal Layers

Panel inputs
    -> operation functions
    -> Graph logic nodes
    -> internal execution runtime
    -> cached Panel outputs

Style

• Use type hints.
• Use structured docstrings for public APIs.
• Follow Black-compatible formatting and an 88-character line length.
• Use logging, not print, in core logic.
• Raise descriptive exceptions and fail fast.

Testing

Add focused tests for:

• Panel validation
• Built-in operation behavior
• User-defined decorated operations
• DAG validation
• Alignment
• Cache reuse
• Hash reuse for already-aligned inputs
• Intermediate Graph.output population
• Graph specification stability