Dyadfit: A Pre-Launch Waitlist and Content Site Built in About 2 Weeks

Overview

Dyadfit is a CodesSavvy product, not a client engagement. It is a pre-launch waitlist and content site for a GLP-1 muscle-preservation companion: a protein target, a strength plan and a weekly check.

The Challenge

The app is not built yet. The site has to explain the idea in English, with key pages in four more languages. Health claims are tied to primary sources and corrected in public when wrong.

Our Solution

The first commit was on 11 Sep 2026 and there were 37 commits to 23 Sep 2026. One route map drives hreflang, the sitemap and the language switcher. Dictionaries are typed. There is no redirect by browser language. A page gets a translated URL only once its text is translated, and a strict test checks for leftover English. Charts are live HTML. Medicines are marked up as Thing, not Drug, after Search Console treated a medicine as a product. Four problems shaped the work. (1) English 'habit cards' shipped on every translated home page past a fully green test suite, because the test only checked strings it was told to look for. We moved the cards into the dictionaries and added the titles to the check list. The same class of bug came back in a later audit, so we added more sentinels. (2) Translated text was clipped inside a fixed-height card in four languages. We found it by measuring scroll size against client size in the browser. (3) A second AI agent, separate from the one that translated, compared each translation line by line against the English source. It caught an invented competitor price, dropped sentences and citations, and label claims credited to the wrong brand. We fixed that last one by naming the molecule, tirzepatide. The review was by AI agents, not by native-speaker humans. (4) Medical claims are tied to primary sources. The lean-mass range moved from 26 to 40 percent (a small case series) to 24 to 39 percent (a meta-analysis of 36 trials with 5,316 people). Sources are listed on Dyadfit's methodology page. The site publishes a corrections log and a section on what the evidence does not establish.

Results

Tech Stack