Build Log

The Medina Square — Live Testing Hardening

July 7, 2026

Zeb live-tested the app against real questions and real mobile devices; every bug found (branding leaks, vague fallback copy, sourcing caution scoped wrong, unformatted answers, HTTPS-only voice input) got fixed at the root, and the app was renamed from medina-navigator to medina-square end to end.

Medina SquareBug FixAI PromptingMobile UXArchitecture

spins fixed via live testing

starter prompts refreshed with tested, real questions

new data pattern (sourceScope on Evidence)

Timeline

Start

Continued from prior session

Voice input/output feature work in progress

Early

Voice input + output shipped

Browser-native SpeechRecognition + SpeechSynthesis, no external service

Mid

Mobile bug hunt

allowedDevOrigins fix, HTTPS-only mic limitation surfaced and explained to the user

Mid-late

Sourcing Anvil

ChatGPT comparison exposed a real gap — sourceScope field added to Evidence, scales beyond one resource

Late

Starter prompts rebuilt

All 7 replaced one at a time with real, live-tested questions

Renamed medina-navigator → medina-square

Full path fix across package.json, vercel.json, root workspace config, lockfile regen

What shipped

Voice input (speech-to-text) wired into GuidedFlow.tsx and AskForm.tsx via a shared useSpeechToText hook — browser-native Web Speech API, zero external dependency

Voice output (read-aloud) for Plan results via useSpeechOutput hook and a 'Read aloud' button on PlanView.tsx

tel: and mailto: links for every phone number and email across ResourceCard.tsx and PlanView.tsx — mobile users can tap to call/email directly

MarkdownLite.tsx — small dependency-free markdown renderer (bold + lists) for Evidence answers, replacing a single-paragraph dump

New sourceScope field ('hyperlocal' | 'standardized_program') on the Evidence type — required, TypeScript-enforced across all 7 existing records

Real, sourced SNAP/Medicaid document checklist Evidence record, citing Ohio Benefits' official 'Things You Need To Apply' page — closes the gap exposed by a ChatGPT comparison

7 starter prompts on /ask fully rebuilt with real, live-tested questions covering utilities, senior services, housing crisis, permits, employer/workforce, and financial hardship

Renamed the entire app from medina-navigator to medina-square: folder, package.json name, vercel.json build command, root workspace registration, all script names (dev/build/deploy:medina-square), console.error log labels, regenerated package-lock.json

“ok, this is not a good look for us... how are we going to handle this at scale?”
— Zeb, after pasting a ChatGPT answer that outperformed The Medina Square's SNAP/Medicaid response — the question that triggered the sourceScope architecture fix