Execution Model

Overview

Graphs define what should be computed. Calling Graph.compute() materializes output panels and caches intermediate results during execution.

signal.compute()
panel = signal.output

Pipeline

Graph construction
    -> validation
    -> dependency resolution
    -> panel alignment
    -> node evaluation
    -> Panel output creation
    -> cache storage or reuse
    -> Graph.output population

Intermediate Outputs

Executing a downstream graph evaluates its dependencies. Every evaluated derived node receives a panel output:

signal.compute()
prediction_panel = prediction.output
signal_panel = signal.output

Current Semantics

• Execution is deterministic.
• Panels are immutable from the public API.
• Multi-input frames align on intersecting indexes and columns by default.
• Intermediate cache values are panels.
• Shared DAG nodes are evaluated once per runtime invocation.
• Stored panel hashes are reused when alignment does not change an input frame.
• Scheduling is sequential.

Parallel scheduling, persisted caches, and explicit invalidation remain future extensions.