docs(workplans): add pdf read adapter plan

This commit is contained in:
2026-05-14 23:17:45 +02:00
parent 63518863aa
commit 3deb728375

View File

@@ -0,0 +1,220 @@
---
id: MKTF-WP-0002
type: workplan
title: "PDF Read Adapter"
domain: markitect
status: todo
owner: markitect-filter
topic_slug: markitect
planning_priority: P1
planning_order: 20
depends_on_workplans:
- MKTF-WP-0001
related_workplans:
- MKTT-WP-0018
created: "2026-05-14"
updated: "2026-05-14"
state_hub_workstream_id: "7445fe6b-f1a9-4383-8053-4337337dc095"
---
# MKTF-WP-0002: PDF Read Adapter
## Purpose
Implement the second concrete `markitect-filter` source adapter:
`source.pdf`, a read-only PDF adapter that satisfies the `markitect-tool`
source adapter contract.
The contract dependency is cross-repo and is tracked as related work rather
than a same-repo State Hub dependency edge: `markitect-tool` `MKTT-WP-0018`.
The first PDF slice should target deterministic text extraction from
digitally-readable PDFs. It should preserve page-level provenance and make
extraction uncertainty visible through diagnostics and quality signals.
## Planned Scope
- Optional PDF dependency profile isolated behind a `pdf` extra.
- Entry point group registration:
`markitect_tool.source_adapters`.
- Lightweight `pdf_adapter_descriptor`.
- Adapter id `source.pdf` with media type `application/pdf` and extension
`.pdf`.
- Inspection for basic PDF metadata, page count, encryption status, and
extractability signals.
- Read-only page text extraction into ordered Markdown segments.
- Page-aware source provenance with source paths, page numbers, page labels
where available, and stable segment ids.
- Configurable first-slice options such as page range, page break markers, and
whitespace normalization policy.
- Structured diagnostics for malformed PDFs, encrypted PDFs, image-only or
scanned pages, empty extraction, partial page failures, unsupported embedded
media, and lossy layout/table handling.
- Quality metadata for confidence, lossiness, skipped pages, warning counts,
extraction backend, and page coverage.
- Tests for descriptor shape, matching, inspection, normalization, malformed
inputs, encrypted or non-extractable inputs where fixtures allow, Markitect
API registry use, and entry point shape.
## Non-Goals
- OCR or scanned-document recognition.
- Pixel-perfect layout preservation.
- Table reconstruction beyond plain text and diagnostics.
- Image, figure, annotation, form, signature, or attachment extraction beyond
future metadata/diagnostic hooks.
- PDF writing/export.
- Network fetching.
- External processes or native system services in the first slice.
- Making PDF dependencies mandatory for EPUB3 or other adapters.
## P2.1 - Pin PDF v1 dependency and extraction policy
```task
id: MKTF-WP-0002-T001
status: todo
priority: high
state_hub_task_id: "2ce51bb9-9182-4927-90d1-4c08433b5ddb"
```
Choose the first PDF extraction backend and dependency profile.
The decision should document:
- pure-Python preference for the first slice
- optional dependency placement under the `pdf` extra
- supported inputs: local, digitally-readable PDFs
- unsupported inputs: scanned/image-only PDFs without OCR
- encrypted/permission-restricted PDF behavior
- how page range, page breaks, and whitespace normalization should behave
- fallback or future status for heavier layout/OCR backends
Output: dependency decision, option contract, and implementation notes.
## P2.2 - Add descriptor and entry point registration
```task
id: MKTF-WP-0002-T002
status: todo
priority: high
state_hub_task_id: "27d754a9-59ae-4419-946b-f1f847bd3b10"
```
Add a `pdf_adapter_descriptor` matching the existing EPUB3 descriptor pattern.
The descriptor should define:
- adapter id `source.pdf`
- version `1`
- media type `application/pdf`
- extension `.pdf`
- read operation only
- safety metadata with local reads only
- option schema for page range, page breaks, and whitespace normalization
- quality profile and dependency metadata
- lazy factory import for the PDF adapter implementation
Output: descriptor, entry point registration, and descriptor tests.
## P2.3 - Implement PDF inspection
```task
id: MKTF-WP-0002-T003
status: todo
priority: high
state_hub_task_id: "33b594e6-d12a-46d5-bc50-6ec1aebaaf65"
```
Implement `inspect` for PDF assets.
Inspection should report:
- title, creators/authors, subject, keywords, producer, creation/modification
dates where available
- page count
- encryption or permission status
- basic extractability signals
- diagnostics for malformed, unreadable, encrypted, or unsupported PDFs
Output: inspection implementation and tests with small fixtures.
## P2.4 - Normalize page text into Markitect Markdown
```task
id: MKTF-WP-0002-T004
status: todo
priority: high
state_hub_task_id: "30c0c777-a4e4-43d1-ac24-6a0f84c7b761"
```
Implement `read` for digitally-readable PDFs.
Normalization should:
- iterate pages in deterministic order
- apply page range filtering
- convert extracted text into Markdown-safe segment text
- create one or more ordered segments with stable segment ids
- preserve page-level provenance on every segment
- optionally insert page break markers
- produce a stable document id and cache key through the Markitect source
contract helpers
Output: read implementation and normalization tests.
## P2.5 - Add diagnostics and quality semantics
```task
id: MKTF-WP-0002-T005
status: todo
priority: high
state_hub_task_id: "8b6a190a-350b-4c61-ac4f-1900673a8cd2"
```
Define PDF-specific diagnostics and quality metadata.
The adapter should distinguish:
- malformed PDF
- encrypted or permission-restricted PDF
- no extractable text
- partially failed pages
- scanned/image-only pages
- dropped layout, tables, figures, annotations, or forms
- unsupported embedded resources
Quality should include extraction backend, page coverage, warning count,
skipped pages, lossiness, and confidence.
Output: diagnostic helpers, quality rules, and tests.
## P2.6 - Add fixtures, docs, and validation
```task
id: MKTF-WP-0002-T006
status: todo
priority: medium
state_hub_task_id: "af597160-e189-42be-8479-c6e0f467d238"
```
Add small deterministic PDF fixtures and documentation.
Validation should cover:
- descriptor shape
- media type and extension matching
- metadata inspection
- page text normalization
- malformed or empty extraction behavior
- registry and entry point shape
- `markitect-tool` API use through `inspect_source` and `normalize_source`
Output: tests, README update, and validation command.
## Validation
Run from `markitect-filter`:
```bash
PYTHONPATH=src:/home/worsch/markitect-tool/src python3 -m pytest
```