Why the Statement Workbench Is Static on Purpose

Most “AI in finance” demos hide the interesting part: where the system stops being a calculator and starts inventing narrative. The statement review workbench does the opposite. It is a static Astro page, synthetic fixtures, and TypeScript arithmetic you can open in a browser with no account, no upload, and no API key.

What “static” buys you

The demo ships as Cloudflare Pages assets. There is no Python runtime, no Streamlit session, and no server that accepts a spreadsheet. That is not a scalability claim. It is a control boundary:

• Inputs are invented scenarios chosen on the page.
• Ratios and working-capital figures are computed in the browser from visible values.
• Observations describe the arithmetic. They do not infer causes or recommend actions.
• Missing prior periods and zero denominators stay unavailable instead of being smoothed over.

If you cannot explain a number without a model, you do not yet have a review workflow—you have a chat window.

What this is not

The workbench is not a production close tool, not an LLM evaluation, and not evidence of a customer deployment. The project write-up and source page state those limits on purpose. A real deployment would still need permission-aware data access, retention rules, finance-approved definitions, and audit persistence before any model-assisted narrative layer is introduced.

Why builders should care

Forward-deployed work often starts by making the existing review path legible. Static prototypes are a cheap way to pressure-test that path with stakeholders: what must be deterministic, what must be human-owned, and what—if anything—a model is allowed to draft later. For related reading on how small errors travel through finance artifacts, see Small AI errors can drift into important decisions. Start from the hub if you want the wider path.