vouch¶
Every number in your paper, vouched for by the code that produced it.
Experiments record their results as they run. The paper cites them by key,
\vouch{key}, and never contains a typed number. vouch check proves that every
cited value exists, comes from a run whose code hasn't changed since, and that
every claim still holds. When a cited value changes, vouch shows you the sentences
to re-read. Any number traces back to the function, arguments, seeds, command and
commit that produced it, from the terminal, a local web page, or by clicking it in
the PDF.
It is built for people and for LLM agents writing papers: the correct number is always cheaper to use than a guessed one.
Start the tutorial{ .md-button .md-button--primary } Quickstart{ .md-button }
Install¶
$ pip install -e . # from a clone; Python ≥ 3.10, no dependencies
$ vouch init # finds your paper, writes vouch.toml, copies vouch.sty
The base install is everything, the MCP server included, with no dependencies
(only tomli on Python 3.10). Extras: .[pandas] to record pandas DataFrames;
.[mcp] installs nothing more (vouch mcp needs nothing); .[test] for the test
suite.
The shape of it¶
| Step | What happens |
|---|---|
| Record | Decorate the function that computes your results (@vouch.track), or call vouch.record/vouch.record_all — every call is recorded with its arguments, seeds, timing and the exact code that ran. |
| Find the key | vouch explore --open browses every recorded value in a local page; vouch search/vouch cite do the same from the terminal. |
| Cite | \vouch{key} in the LaTeX source. vouch build renders it; latexmk compiles as usual. |
| Compute | Differences, ratios and derived tables live in vouch_values.py, evaluated by vouch build — never in anyone's head. |
| Check | vouch check proves every cited value exists, is fresh, and every claim still holds — in under a second, with no experiment re-run. |
Why¶
Numbers in papers go wrong in four ways: they're typed wrong, never produced, out of date after the code changed, or the number was updated correctly but the prose around it ("the best of all models") is now false. vouch is built to close all four, for a human author and an LLM agent alike — see Concepts for the full motivation.
Looking for exhaustive, implementation-level detail? The full design spec in the repository is authoritative; this site is a guided, web-friendly tour of the same material.