CADS Workbench

English | 简体中文

Corpus-Assisted Discourse Studies Research Workspace

CADS Workbench is a Windows desktop application for corpus-assisted discourse studies: import a news corpus, run reproducible language analyses, review candidates by hand, organize evidence into claims, and write it up with full provenance — from one application, without installing Python.

Windows x64 · v1.0.0-rc1 (Release Candidate) · MIT License · No Python required

Download Latest RC · Quick Start · Release Notes · 中文说明

Download: CADS-Workbench-1.0.0-rc1-win-x64.zip — extract fully and run CADS Workbench.exe.

---

Research Workflow

Corpus → Analysis → Review → Evidence → Claim → Writing → Run Compare

Every analytical result is published as an immutable Published Run with a full manifest (corpus fingerprint, parameters, file hashes). Evidence binds to that generation, so any number in your writing can still be traced to its source years later.

Why CADS Workbench

• It does not just produce statistics tables — automated results are candidate evidence, never final conclusions
• Analysis results are frozen into auditable Published Runs
• Evidence is bound to a specific published generation, with captured context snapshots
• Claims and writing stay traceable to the original KWIC lines, patterns, and runs behind them
• Runs can be compared — older evidence is never silently replaced
• Human review (semantic prosody, source/country) is part of the workflow by design

Main Features

• Project Hub — create or open projects, recent-project list, guided new-project wizard (research templates, corpus import, analysis defaults)
• Corpus — LexisNexis DOCX / TXT / CSV / Excel import, document registry, source & country review, corpus health (sanity) gate
• KWIC / Concordance — target-centered concordance with filtering and context inspection
• Collocates — window-based co-occurrence candidates ranked by MI and G²
• Phrases — modifier-phrase and lexical-bundle candidates
• Group Comparison — compare pattern distributions across source, institution, country, or custom groups
• Source / Country Review — human review of normalization and inferred country labels, with the evidence shown inline
• Semantic Prosody Review — keyboard-driven coding workbench for semantic-prosody candidates
• Published Runs — auditable execution history with cancel, crash recovery, and transactional publication
• Evidence Trail — capture KWIC / patterns as evidence bound to the published generation it came from
• Claims — organize evidence into research claims (multi-reference, integrity-checked, never auto-judged)
• Writing — sectioned workspace with claim/evidence reference cards
• Run Compare — compatibility-gated comparison of two published runs
• Markdown Export — research draft with evidence appendix (draft and clean modes)

Quick Start

1. Download and fully extract CADS-Workbench-<version>-win-x64.zip
2. Double-click CADS Workbench.exe (no Python installation needed)
3. New Project — follow the four-step wizard (name, research template, corpus, target terms)
4. Import corpus — LexisNexis DOCX, TXT folder, or CSV/Excel
5. Run the sanity check, then New Analysis Run
6. Review coded candidates, collect Evidence, organize into Claims
7. Write with live evidence cards and export a Markdown draft
8. Close and reopen anytime — projects and recent list are restored

A full user guide is in docs/QUICKSTART_GUI.md. Prefer the command line? The classic CLI remains available — see docs/WORKFLOW.md.

Data Safety & Reproducibility

• Analysis execution and publication are separated: a new result becomes an official Published Run only after the publication transaction commits — failures roll back completely, never "half new, half old"
• Evidence is bound to immutable published generations and traceable to corpus fingerprint, analysis parameters, publication manifest, and document/KWIC/pattern provenance
• Automated outputs are candidate evidence, not final interpretation; human review is part of the workflow by design
• No telemetry, no analytics, no crash uploads. Diagnostics are generated only on explicit user export and exclude corpus text and research content
• The application never modifies your corpus; all writes have defined boundaries with atomic saves and rollback
• This software does not replace system backups — keep your own project backups

Methodological Boundaries

• MI / G² are used for pattern discovery and ranking — they do not by themselves constitute significance proofs or discourse conclusions
• Semantic prosody candidates require human review
• Source/country inference is an auxiliary variable and must be confirmed manually
• Automated statistics cannot replace close reading of the context
• CADS Workbench does not generate substantive discourse-research conclusions automatically

These boundaries follow the methodology documentation (docs/METHODOLOGY.md) and are repeated in the Release Notes.

Known Limitations

See Release Notes for the full list. Highlights:

• Primarily designed for English corpora
• NLP model: spaCy en_core_web_sm (small model accuracy bounds)
• Writing export is Markdown; no DOCX/PDF yet
• No AI-assisted writing or automatic conclusions
• One project per window; no file associations; distributed as a directory ZIP (no installer yet)

Repository Layout (for contributors)

Path	Purpose
gui_next/	Current PySide6 research workbench (the CADS Workbench product)
shared/, modules/, LexisWordToTxt/	Frozen analysis pipeline (import, extraction, grouping, review artifacts)
research_tool.py	Unified CLI (cads)
pipeline.py	Legacy compatibility CLI
integrated_app.py	Legacy Tkinter GUI (cads-gui-legacy)
docs/	Product, methodology, and engineering documentation
tests/	Test suite (unit, GUI, E2E)

Build instructions: docs/BUILDING.md. Release checklist: docs/RELEASE_CHECKLIST_V1.md.

License

MIT — see LICENSE.