Blake Corpus Analysis Platform

Blake builds a reproducible local corpus workflow for William Blake texts, images, metadata, analysis results, visual summaries, exports, and reports. The API is BlakeCorpus.

Primary Workflow

uv sync
uv run blake corpus run --quick --output-dir analysis_output
uv run blake corpus run --skip-download --output-dir analysis_output --fixture-mode never --viz-profile core --format json --archive-scope archive-inventory --image-mode archive-all --image-dpi 300 --max-images-per-work none
uv run blake corpus public-release-preflight --output-dir analysis_output --data-dir blake_data
uv run blake corpus public-release-package --output-dir analysis_output --data-dir blake_data

A quick run is offline and uses the local blake_data/ cache. If that cache is empty, quick mode seeds a tiny bundled fixture corpus so the pipeline proves non-empty processing, visualization, export, and reporting without network access. A full run enables live acquisition through registered source providers.

For manuscript/regression work, the canonical local regeneration command is the second command above: it reuses the existing local corpus with --skip-download, then regenerates metadata, analysis, visualizations, exports, and reports. Do not add --skip-analysis unless you are explicitly testing a stale-analysis scenario.

archive-all is an evidence-depth setting for local Archive image acquisition. It is not a redistribution permission. Public-release preflight and package commands stage metadata, checksums, and release-safe files locally; they do not publish, tag, deposit, or clear rights-gated source materials.

Output Layout

All corpus artifacts are written under analysis_output/ by default:

analysis_output/
├── pipeline_summary.json
├── corpus_manifest.json
├── data/
│   ├── source_discovery.json
│   ├── source_discovery_health.json
│   ├── source_audit.json
│   └── evidence_acquisition_review.json
├── analysis/
│   ├── analysis_summary.json
│   ├── analysis_diagnostics.json
│   └── analysis_results.json
├── visualizations/
│   ├── visualization_summary.json
│   ├── target_coverage_matrix.json
│   ├── target_coverage_matrix.png
│   └── corpus_workbench.html
├── exports/
│   └── export_summary.json
├── reports/
    ├── completeness_report.json
    ├── source_coverage.json
    ├── provenance_manifest.json
    ├── chronology.json
    ├── theme_entity_network.json
    ├── text_image_linkage.json
    ├── statistics.json
    ├── private_release_preflight.json
    ├── private_release_preflight.md
    ├── public_release_preflight.json
    ├── public_release_preflight.md
    ├── rights_gate_report.json
    ├── rights_gate_report.md
    └── corpus_report.md
└── releases/
    └── public_release/
        ├── public_release_manifest.json
        ├── public_release_bundle.zip
        ├── CITATION.cff
        └── zenodo_metadata.json

Raw downloads, SQLite metadata, cached texts, and images live under blake_data/.

Python API

import asyncio
from pathlib import Path

from blake import BlakeCorpus, CorpusConfig

async def main():
    corpus = BlakeCorpus(Path("blake_data"), output_dir=Path("analysis_output"))
    summary = await corpus.run(CorpusConfig(quick=True))
    print(summary.status)

asyncio.run(main())

CLI

blake corpus run --quick --output-dir analysis_output
blake corpus run --quick --viz-profile full
blake corpus status --coverage
blake corpus visualize --profile core
blake corpus run --format json --format csv --format markdown
blake corpus public-release-preflight --output-dir analysis_output --data-dir blake_data
blake corpus public-release-package --output-dir analysis_output --data-dir blake_data
blake search --query "tyger"
blake info --work-id <id>
blake explore

The corpus workflow lives under blake corpus; search, info, and explore are convenience commands over local corpus metadata.

Current Evidence Contract

• analysis_output/corpus_manifest.json is the target-ledger coverage contract.
• analysis_output/data/source_audit.json records source discovery, provider health, URL checks, Archive metadata enrichment, and missing-evidence candidates.
• analysis_output/data/evidence_acquisition_review.json is the strict review ledger for unresolved targets and manual source leads.
• analysis_output/analysis/analysis_summary.json and analysis_output/analysis/analysis_diagnostics.json record current analysis status, eligibility, completion, diagnostics, and errors.
• output/figures/figure_registry.json records every manuscript figure, its brief, accessibility notes, statistical question, and generated statistics.
• analysis_output/reports/private_release_preflight.* and analysis_output/reports/rights_gate_report.* are local-only release/readiness checks. analysis_output/reports/public_release_preflight.* and analysis_output/releases/public_release/ are the public GitHub + Zenodo release staging surfaces; they generate bundle metadata, checksums, and citation files without altering the repository itself.

Corpus Components

• blake.corpus owns CorpusConfig, CorpusManifest, CorpusRunSummary, PhaseResult, source registry, and BlakeCorpus.
• The corpus manifest compares local works against a versioned canonical target ledger and reports present, partial, missing, duplicate, and unverified coverage.
• blake.downloader keeps legacy adapters while corpus acquisition normalizes source records through SourceProvider wrappers.
• blake.metadata stores and searches local work metadata.
• blake.analysis writes typed analysis records through the corpus pipeline.
• blake.viz exposes one Pydantic visualization configuration model.
• blake.export supports JSON, CSV, Markdown, HTML, PDF, and Parquet where dependencies permit.

Data Sources

Default source adapters cover Blake Archive, Project Gutenberg, and Internet Archive. Additional museum and discovery adapters exist for British Museum, Tate, Wikimedia, HathiTrust, and related sources. Live-source integration is optional; the default test and quick-run surfaces do not require network access. A run is only reported as complete when the canonical target ledger is fully covered; otherwise reports explain the remaining gaps.

Development

uv sync
uv sync --extra dev --extra api
uv run pytest tests/ -q
uv run pytest tests/ -q -m "not slow and not integration"
uv run blake corpus run --quick --output-dir analysis_output
uv run blake corpus run --skip-download --output-dir analysis_output --fixture-mode never --viz-profile core --format json --archive-scope archive-inventory --image-mode archive-all --image-dpi 300 --max-images-per-work none
uv run blake corpus private-release-preflight --output-dir analysis_output --data-dir blake_data
uv run blake corpus public-release-preflight --output-dir analysis_output --data-dir blake_data
uv run blake corpus public-release-package --output-dir analysis_output --data-dir blake_data

The default install keeps GUI/API serving packages out of the core dependency set. Use --extra gui for Streamlit/Gradio surfaces and --extra api for FastAPI serving.

License and Data Rights

The project code and project-authored documentation are licensed under the MIT License. See LICENSE.

Downloaded or cached Blake materials under blake_data/, generated image mosaics, source TEI, transcriptions, and provider-supplied images are not relicensed by this repository. Treat them under their source-provider terms and the manifest's recorded authority/license metadata.

The manuscript's legal distribution section and rights-matrix supplement are the release checklist for public artifacts. In short: publish code, commands, checksums, source URLs, manifests, and aggregate diagnostics; keep raw provider files and local evidence caches out of public source releases unless permission, source-compatible licensing, or a venue-specific fair-use rationale is documented.

uv run blake corpus private-release-preflight writes local-only preflight and rights-gate reports under analysis_output/reports/. It never publishes, tags, deposits, or redistributes artifacts; it records what a private GitHub-oriented review would include and which public artifact classes remain blocked.

uv run blake corpus public-release-preflight and uv run blake corpus public-release-package prepare the GitHub + Zenodo release bundle under analysis_output/releases/public_release/. The package includes code, safe generated outputs, CITATION.cff, Zenodo metadata, and SHA-256 checksums while excluding the local manuscript pipeline, raw source caches, and rights-gated assets by default.

Citation and Archive

This project is archived on Zenodo with a DOI, cross-linked to this repository:

• DOI: 10.5281/zenodo.21047574
• Zenodo record: the working-paper PDF (rights-safe; Blake Archive image mosaics omitted) plus the open-source software release bundle.
• See CITATION.cff for citation metadata.

The MIT license covers project code and project-authored outputs only. Provider-supplied Blake Archive TEI, transcriptions, images, and fallback source texts are not included here and remain under their source-provider terms.