notarius/docs/internal/pipeline.md

Pipeline Internals

The implemented pipeline runner lives in internal/framework/pipeline. It executes the fixed workflow defined by the architecture policy:

input -> chunk -> extract -> merge -> normalize -> output

Pipeline execution is serial. The runner executes the resolved lanes one after another in the fixed workflow order.

Profile Resolution

Config loading produces pipeline.PipelineProfile values. Resolution happens before execution:

1. internal/core/config.Config.Resolve validates config and finds the named pipeline.
2. The optional lane selection is passed to pipeline.ResolvePipeline.
3. Module bindings are defaulted:
   • chunk: generic
   • merge: appendorder
   • normalize: noop
   • output: json
   • LLM profile: empty, which lets Scriptorium prompt defaults choose a profile.
4. The module catalog is checked for each bound module key.
5. Module capabilities are checked in workflow order.
6. A digest is calculated from the resolved pipeline without the digest field.

The CLI writes the resolved pipeline and digest to diagnostics.

Pipeline profiles and artifact lanes may include reference binding maps keyed by reference slot name. During resolution, pipeline-level bindings act as defaults for selected chunk, extractor, merger, and normalizer targets that declare the slot; target-local bindings override or add bindings for that target. Runtime --reference requests override target config bindings, and runtime unbinds remove optional target bindings. Flat runtime slot names are resolved only when exactly one selected target declares the slot; otherwise the CLI requires a more specific selector such as chunk.slot, lane.extract.slot, lane.merge.slot, or lane.normalize.slot. Resolution validates bindings against the declaring target specs and stores the bindings in target-aware resolved reference holders. It does not read reference files or include reference bytes in source digests.

During run preparation, resolved file references for chunk, extractor, merger, and normalizer targets are materialized before any LLM-backed pipeline work. Config bindings resolve relative to the config file, and CLI bindings resolve relative to the current working directory. Materialization accepts UTF-8 text files, computes sha256: content digests, records file origins, infers canonical base media types from file extensions, enforces declared byte limits, and warns for empty bound files. Media-type acceptance is checked only when a slot declares AcceptedMediaTypes; unknown extensions are recorded as application/octet-stream. Reference content is omitted from diagnostics and manifests. The CLI writes provenance-only resolved reference diagnostics, and the run manifest records target-stage reference provenance separately from source digests. Runtime reference content is passed to the matching chunker, extractor, merger, or normalizer request. LLM-backed modules pass that material onward as named Scriptorium prompt inputs.

The CLI carries raw input bytes into pipeline.RunInput. Input adapters parse those bytes into the source document, while LLM-backed modules that need the original transcript material can pass the same bytes as a prompt input with origin metadata. The raw input payload is not written to manifests or default diagnostics.

The CLI also carries an optional run session_id. The runner makes it available to chunk, extract, merge, and normalize requests; LLM-backed modules forward it through their structured completion requests so Scriptorium can include it in prompt execution metadata.

Registries And Module Specs

pipeline.Registries holds concrete constructors for execution. A pipeline.ModuleCatalog exposes module specs for config validation and resolution.

Every production module registers a ModuleSpec with:

• Key: module key used in config;
• Stage: module kind such as input, chunk, extract, merge, normalize, validate, or output;
• Provides: capabilities added after that module runs;
• Requires: capabilities that must already be available.

Chunk, extract, merge, and normalize specs may also declare reference slots. Slot declarations are available from registry metadata without constructing module instances. Input, validate, and output specs must not declare reference slots.

Capability checks prevent incompatible pipeline composition before a run starts.

Runner Input And Output

pipeline.RunInput carries:

• a ResolvedPipeline;
• optional source ID, input path, and raw input bytes;
• a structured LLM client;
• run ID, start time, LLM profile manifest metadata, and CLI metadata.

pipeline.RunOutput carries:

• run manifest;
• normalized raw outputs;
• rejected raw outputs;
• warnings;
• logical output files returned by the output encoder.

The CLI owns durable file writes and diagnostics writes after the runner returns.

Execution

The runner:

1. validates run input and registries;
2. builds the input adapter and parses the raw input into a source document;
3. validates the source document;
4. builds the chunker and produces source chunks, retrying when configured;
5. validates source chunks against framework invariants and any registered raw chunk validators;
6. runs each selected artifact lane in sorted resolved order;
7. builds the output encoder and validates logical output file names.

Chunk Results

Chunkers implement contracts.Chunker and receive a contracts.ChunkRequest with the validated source document, reference set, structured LLM client, the configured LLM profile, module options, and run metadata. Deterministic and LLM-backed chunkers use the same contract; provider construction stays outside chunk modules.

When chunking succeeds, the runner validates generic chunk invariants before running extractors:

• chunk IDs must be non-empty and unique in the chunk result;
• each chunk SourceID must match the source document ID;
• each chunk Index must match its zero-based returned order;
• each chunk start and end unit ID must exist in the source document, with the start unit at or before the end unit;
• each chunk must include non-empty extraction content and media type;
• each chunk must contain at least one source unit;
• a chunk must not repeat a source unit;
• every chunk source unit must exist in the source document;
• source units inside each chunk must appear in source-document order.

After validation, the runner rebuilds each chunk from source-document units by integer ID, preserving chunk boundaries, content bytes, media type, and cloned chunk metadata. Extractors and downstream stages therefore see canonical source units, while SourceChunk.Metadata remains the supported place for chunker-owned context.

If chunk validation rejects a chunk after configured retries, the runner records a rejected raw output and skips downstream lane execution. Framework-level chunking or validation errors that remain after configured retries fail the run.

The framework does not require complete source-unit coverage and does not reject overlap between different chunks. Stricter policies, such as full coverage or non-overlap, belong to individual chunk modules when they are part of that module's contract.

Within an artifact lane, the runner:

1. builds the extractor, merger, and normalizer;
2. records module manifest metadata when modules provide it;
3. extracts one raw ExtractOutput from each accepted chunk, retrying when configured;
4. fills runner-owned provenance on each extract output, including lane ID, extractor key, source ID, chunk ID, and chunk index;
5. validates raw extract outputs and omits rejected outputs from merge input;
6. merges ordered accepted extract outputs into one raw MergeOutput, retrying when configured;
7. validates raw merge output and skips normalization for rejected merge output;
8. normalizes the accepted merge output into one raw NormalizeOutput, retrying when configured;
9. validates raw normalize output and appends accepted normalized raw output to RunOutput.NormalizeOutputs.

Validators

The current runner handoff is raw-output based. Extractors, mergers, and normalizers do not advertise candidate validator chains through their module interfaces. Runner-side raw validation chains receive the raw module output plus stage, lane, module, source, and chunk provenance. Empty raw validation chains approve output by default.

Validator rejection is a non-fatal run outcome: the rejected output is recorded in RunOutput.Rejected and does not pass to the next stage. Validator execution errors are framework-level errors and retry according to the relevant binding.

Warnings And Failures

Warnings from the successful chunking, extraction, merging, and normalization attempts whose outputs are used are accumulated in RunOutput.Warnings, along with output encoder warnings. Warnings from discarded retry attempts are not promoted to final warnings.

Errors wrap the operation and module key or lane context. If execution fails after a manifest exists, the returned manifest is marked failed and receives a completion timestamp.

On successful execution, the manifest validation status is:

• approved when no raw outputs were rejected;
• rejected when at least one raw output was rejected.

Manifest Population

The manifest records run ID, pipeline ID, pipeline digest, module keys, top-level module metadata, artifact lanes, LLM profile metadata, source digest, reference provenance, validation status, and timing.

Singleton pipeline modules may add non-secret metadata by implementing contracts.ManifestMetadataProvider. The runner records that metadata under module_metadata with stable keys for input, chunker, and output.

Lane-owned modules may add non-secret metadata through artifact_lanes[].metadata. The runner records extractor, merger, and normalizer metadata there. The D&D spell extractor uses lane metadata for prompt and response-schema provenance.