Seasonal Pricing Console

A seasonal pricing console for short-term rental hosts. It runs entirely in the browser, no backend and no API key, and answers pricing questions with fixed, read-only SQL rather than a chatbot's free text.

The point is not another chat window. A host needs the profit-maximizing nightly rate for each of 15 Ipoh listings by season, and a reason to trust it. The app is the harness around that, not a wrapper on a model.

The console has four screens: Overview, Price Sensitivity, Recommended Rates, and The Pricing Questions. The brain underneath is a difference-in-differences elasticity estimator with listing fixed effects and a weekend control, built and proven before any screen was styled. On data with a baked-in known elasticity, the estimate recovers it with a mean error of 0.224 against 0.713 for a naive read, inside the 95% interval in all 12 type-season cells.

The Pricing Questions tab is the governed layer. Ten answers come from the locked results; four run live as fixed sql.js queries over a small rates table inside the browser:

• The query layer picks an approved track and binds values only, never authoring its own SQL.
• The hidden true elasticity stays in an eval-only column the queries can never reach.
• Every figure is computed, not generated, so the same question returns the same number every time.

The moat is the governed query layer and the recovery check, the two things a chat model or NotebookLM cannot do over tabular data. Strip those and it collapses into a wrapper.

The result is 15 listings priced, about RM254,138 in projected annual uplift, and a console a host can open and act on with no key to leak.