Module bundle 

SafeBundle generation: OCR + Gaze redact → on-disk artifacts.

The top-level clean function is the public adopter entry point. It routes any supported input (PNG / JPG / single-page PDF) through OCR, pipes the extracted text through a gaze::Pipeline, and persists the result as three files split across agent and owner target directories:

agent_out/
  clean.md        # OCR text with PII replaced by reversible tokens
  report.json     # BundleReport — OCR + PII counts + provenance

owner_out/
  manifest.json   # gaze::Manifest — restorable, canonical

The manifest contract is the same one the rest of the gaze runtime uses (gaze::Manifest). Because it carries restore material, it is written only to the owner output directory.

Structs

AgentBundleDir

Agent-visible SafeBundle output directory.

BundleReport

Bundle audit + provenance report serialized to report.json.

ClassCount

Per-class PII detection count for BundleReport.

LayoutSummary

Opaque layout summary placeholder.

OwnerBundleDir

Owner-only SafeBundle output directory.

PageReport

Per-page OCR/layout provenance.

Pipeline

Configurable document-cleaning pipeline.

SafeBundle

Post-ingestion artifact paired with a Gaze Manifest.

Enums

OcrSource

Per-page extraction source.

Constants

BUNDLE_VERSION

Versioned report.json schema tag (bump on breaking shape changes).

CLEAN_MARKDOWN_FILE

Bundle filename written into the agent output directory for tokenized Markdown.

MANIFEST_FILE

Bundle filename written into the owner output directory for the restorable manifest.

REPORT_FILE

Bundle filename written into the agent output directory for the OCR + PII provenance report.

Functions

cleanocr-tesseract

Top-level entry point: ingest one document, write a SafeBundle to disk.

clean_with_ocr_backendocr-tesseract

Top-level entry point with an adopter-supplied OCR backend.