Super Converter V2

Super Converter V2 is the executable work unit for a new local-first file conversion product. The repository now contains both the Engine Lab and the first desktop workbench shell, rather than extending Liquid Converter V1 in place.

Boundary

• This repo is the implementation home for Super Converter V2
• It is not a continuation of the silver-planck runtime layout
• silver-planck remains the V1 reference repo and the source of first-batch fixtures and planning assets
• The default architecture excludes LibreOffice
• Rust + Tauri remains the preferred product host direction

Current Scope

This repo currently proves:

• Rust workspace boundary
• fixture corpus and fixture manifest generation
• capability registry
• preflight analysis
• normalized adapter contract
• bake-off harness
• first-round engine recommendations
• Tauri desktop shell with React/Vite workbench
• local desktop job model for Convert, Queue, History, Diagnostics, and Settings

It does not yet ship OCR, Keep Layout, LibreOffice-backed defaults, or rich document preview.

Workspace Layout

super-converter/
├── apps/
│   ├── cli/
│   └── desktop/
├── crates/
│   ├── orchestrator/
│   ├── capability-registry/
│   ├── engine-adapters/
│   ├── artifact-bundle/
│   ├── diagnostics/
│   └── preflight/
├── fixtures/
├── benches/
├── scripts/
└── docs/

Default Candidates

• pdfium-render as the first PDF native candidate
• calamine as the first XLSX structured candidate
• MarkItDown, Docling, and Pandoc via sidecar or subprocess adapters

Default Non-Goals

• no LibreOffice in the default path
• no Apache Tika in the default path
• no desktop UI in Phase 1A
• no free-form any-input to any-output matrix

Quick Start

cargo run -p cli -- fixtures manifest
cargo run -p cli -- preflight run fixtures/test_pdf.pdf
cargo run -p cli -- doctor docling
cargo run -p cli -- bakeoff run-all
cargo run -p cli -- bakeoff summarize
cd apps/desktop && npm install
cd apps/desktop && npm run tauri dev

Docling Runtime

• docling stabilization in this repo assumes the .venv312 runtime
• on macOS, Homebrew spatialindex is the canonical fix when libspatialindex_c is missing from the local runtime
• use cargo run -p cli -- doctor docling to verify readiness before trusting non-PDF deep-extract results
• PDF deep extraction via docling is intentionally excluded from the current default architecture

Desktop Workbench

• apps/desktop is the Phase 2A industrial workbench shell
• the shell surfaces Convert, Queue, History, Diagnostics, and Settings
• pdfium-render remains the default PDF path
• docling is exposed only as the non-PDF deep-extract path