pytauargus-tui

An interactive terminal UI for Tau-Argus, the statistical disclosure control engine. It is a thin view over the pytauargus engine; every action is a call the headless CLI already makes. Installing it also gives you the tauargus CLI.

Screenshots

What it does

• Runs supported .arb batches. It runs the engine pipeline (apriori safety rules, suppression, rounding, audit), and every computed table is browsable. Start without a file and a .arb file browser opens.
• Colours cells by status. Green safe, red unsafe (a sensitive value), cyan protected, yellow hidden by a suppression, dim empty. All 14 engine status codes are recognised and named.
• Shows what any cell means. The detail pane shows the value, status name, pending mark, and, after an audit, the realised feasibility interval for the cell under the cursor.
• Lets you mark cells before solving. u requires suppression, p protects (keeps), S overrides safe. Marks are instant draft edits with toggle/replace and undo/redo; the solver never runs until you ask.
• Previews suppression and rounding. s (OPT, whole table) and r (RND) run on a throwaway copy of your batch. You inspect the previewed grid, then y accepts or n cancels. Your working table never changes behind your back, and a preview that is no longer current is flagged.
• Audits. Realised feasibility intervals per cell (a).
• Exports. Writes the accepted table as a legacy INTERMEDIATE .tab file (o).
• Saves the workspace. A versioned, fingerprinted JSON manifest of the accepted operations and pending cell marks, written next to the .arb (w).
• Reopens a saved workspace. Launch tauargus-tui on the .workspace.json, or pick it in the file browser. The accepted result is rebuilt through the engine, the pending draft comes back (undoable), and the revisions are continuous. If the .arb changed since the save, the old outcome is not treated as current; the current input opens fresh instead, with a clear message.
• Writes a replay batch. R materialises the accepted recipe as a headless-runnable .arb (plus apriori files) next to the input. tauargus run <stem>.replay.arb replays the recorded operator intent. It reruns the solver and does not restore or guarantee an identical numerical result.

What it can't do (yet)

• Inspect all dimensions. The grid uses the first two variables and has no selector for higher-dimensional slices. Use the Python/CLI workflow for higher-dimensional inspection; the grid is not a complete view.

• Restore the exact accepted result. Reopening reruns the recipe. The manifest hashes the .arb, not referenced data, metadata, hierarchy or apriori files; changes in those dependencies are not detected by that hash. Keep the complete inputs and accepted outputs when reproducibility matters.

• Provide a publication-output workflow. o exports a diagnostic INTERMEDIATE .tab, not a masked release table. The grid shows original responses, including after rounding. Inspect actual rounded values and select publication output explicitly through the engine API/CLI.

• Offer every legacy method. The headless engine rejects GH/NET/CTA; the interactive method shortcuts are OPT and RND.

• Mark total cells. That needs the real total code, which the display labels do not preserve; the app blocks such marks with a message.

• Choose the rounding base. r always uses the engine's minimum base.

• Work on several batches. One .arb per session, and the batch spec itself is not editable from the UI.

Install

pip install pytauargus-tui
# or, as an extra of the engine package:
pip install "pytauargus[tui]"

Run

tauargus-tui                          # opens a file browser (.arb + saved workspaces)
tauargus-tui path/to/batch.arb        # opens that .arb batch directly
tauargus-tui path/to/batch.workspace.json   # reopens a saved workspace

The full user guide covers every key, every status code and the whole workflow: docs/tui.md.

Develop (monorepo)

cd bindings/tui
uv sync --extra test
.venv/bin/python -m pytest