Module pipeline 

Context Pipeline

The pipeline defines the processing stages that content flows through between raw input and the compressed output delivered to the LLM.

Pipeline Flow

Input → Intent → Relevance → Compression → Translation → Delivery

• Input: Raw file content / shell output enters the pipeline
• Intent: Task-conditioned filtering — what is relevant to the current goal?
• Relevance: Graph/heatmap-based prioritization of content sections
• Compression: AST signatures, entropy filtering, delta encoding
• Translation: Token shorthand (TDD), symbol replacement
• Delivery: LITM positioning, CRP formatting, final output assembly

Each layer can be enabled/disabled per profile (see core::profiles). PipelineStats aggregates per-layer metrics across all runs for observability.

Structs

AggregatedMetrics

Cumulative token counts and timing for a single pipeline layer across all runs.

LayerInput

Content and metadata passed into a pipeline layer for processing.

LayerMetrics

Performance metrics for a single layer execution: tokens in/out, timing, ratio.

LayerOutput

Result produced by a pipeline layer after processing.

Pipeline

A chain of processing layers that content flows through sequentially.

PipelineStats

Persistent aggregated statistics across all pipeline runs.

Enums

LayerKind

Identifies a stage in the compression pipeline.

Traits

Layer

A single processing stage in the compression pipeline.

Functions

is_layer_enabled

Returns whether a given layer is enabled according to a profile’s pipeline config.