Multivert

A credible, free, sharable web quiz that classifies takers across all five "vert" types — introvert, extrovert, ambivert, omnivert, and otrovert — using a multi-axis scoring model that respects each type's defining feature. Every taker sees five independent fit percentages with a dominant-type headline, plus links to peer-reviewed and clinical sources.

Built with vibes, AI, and a small pile of cited research. Not a diagnosis. Don't put it on your résumé.

Stack

• SvelteKit (Svelte 5) + Tailwind CSS
• @sveltejs/adapter-cloudflare — deploys to Cloudflare Pages
• TypeScript (strict)
• Vitest + Playwright + Lighthouse CI

Features

• Five "vert" types, not a bucket. Introvert, extrovert, ambivert, omnivert, otrovert — each scored on its own 0–100% scale. The bars don't sum to 100. An introverted otrovert can legitimately score high on both.
• Continuous sliders, not radio buttons. Range -1.0 → 1.0 with a "deliberately neutral" zero. The thumb stays hidden until you move it, so "unset" is visibly different from "I picked the middle."
• Multi-axis scoring with reverse-scored items. 35 questions across four dimensions. 31% are reverse-scored to keep the otherness axis honest. All items rewritten in the project's voice — meaning preserved, IPIP boilerplate not.
• Deterministic engine. Pure TypeScript, fully client-side, unit-tested. The locked weight matrix lives in docs/PRD.md and changes only via PRD revision. AI is permitted for result-paragraph tone, never for score computation.
• Cited sources, no diagnosis. IPIP, Adam Grant, Cleveland Clinic, Kaminski. Surfaced on the results page so you can argue with them.
• No tracking, no persistence. Slider answers never leave the browser. No analytics pixels, no third-party trackers, no cookies you didn't ask for.
• Mobile-first, keyboard-navigable. Big touch targets, arrow keys nudge the slider, Enter advances.

Architecture

Single-scroll SvelteKit app, statically rendered onto Cloudflare Pages. The scoring engine is deterministic and lives entirely client-side — no backend, no database, no telemetry. The release pipeline is the only moving part worth diagramming.

flowchart LR
    PR[PR opened] --> CI[ci.yml<br/>format · lint · typecheck<br/>unit · build · e2e · semgrep · sonar · codeql]
    CI -->|merge| Main[main]
    Main --> RP[release-please.yml]
    RP -->|opens / updates| RPR[Release PR]
    RPR -->|merge| Tag[GitHub Release<br/>+ tag]
    Tag -->|workflow_call: tag| Deploy[deploy-pages.yml<br/>checkout tag · pnpm build]
    Deploy -->|wrangler-action| CF[Cloudflare Pages<br/>production]

Loading

main only deploys when a Release Please PR merges. Direct pushes don't ship — they just queue commits for the next release.

Deploy & CI

Three workflows under .github/workflows/:

Workflow	Trigger	Purpose
ci.yml	push to main, PR, manual	Quality gate: format, lint, typecheck, unit, build, Playwright, secret scan, Sonar
release-please.yml	push to main, manual	Cuts release PRs from Conventional Commits; chains deploy-pages.yml on release_created
deploy-pages.yml	workflow_call(tag) or workflow_dispatch(tag)	Reusable: checks out the tag, builds, deploys to Cloudflare Pages via cloudflare/wrangler-action, smoke-tests the URL

Cloudflare credentials (CLOUDFLARE_API_TOKEN, CLOUDFLARE_ACCOUNT_ID) are scoped to the deploy GitHub environment, not the repo. Lighthouse runs locally via lefthook pre-push, never in CI — runner variance makes the 100/100/100/100 floor too noisy on shared infra.

Need a one-off prod deploy from a specific tag? gh workflow run deploy-pages.yml -f tag=v1.2.3.

Project structure

.
├── src/
│   ├── routes/              # SvelteKit pages — single-scroll quiz + results
│   ├── lib/
│   │   ├── scoring.ts       # deterministic per-archetype fit calculation (do not touch without PRD)
│   │   ├── archetypes.ts    # ideal vectors + weight matrix per type
│   │   ├── descriptions.ts  # result-paragraph copy
│   │   ├── questions.ts     # 35-item question bank with reverse-score flags
│   │   ├── components/      # Svelte 5 components (sliders, chapter intros, result bars)
│   │   └── state/           # rune-based quiz state
│   ├── app.css              # Tailwind v4 entry (@import 'tailwindcss')
│   ├── app.html             # document shell
│   └── app.d.ts             # SvelteKit ambient types
├── e2e/                     # Playwright specs — smoke + flow
├── docs/
│   ├── PRD.md               # locked scoring model + weight matrices
│   └── voice.md             # tone reference for result copy
├── static/                  # public assets
├── .github/workflows/       # ci.yml, release-please.yml, deploy-pages.yml, codeql.yml
├── AGENTS.md                # contributor + AI-agent contract
├── Makefile                 # canonical command surface
├── svelte.config.js         # adapter-cloudflare config
├── vite.config.ts           # Tailwind v4 plugin + dev server
└── release-please-config.json

Getting started

make install   # install dependencies
make dev       # start the dev server
make test      # run unit + integration tests
make e2e       # run Playwright E2E
make build     # production build

See the Makefile for the full target list.

Documentation

• docs/PRD.md — product requirements, scoring model, and locked weight matrices.
• AGENTS.md — repository conventions for AI agents and contributors.

Acknowledgements

The scoring model leans on actual research, not vibes-only:

• IPIP-BFFM Big Five Test (openpsychometrics) — extraversion construct baseline.
• Adam Grant — Rethinking the Extraverted Sales Ideal (Wharton, 2013) — empirical case for ambiversion as a distinct, peak-performing third pole.
• Cleveland Clinic — Omnivert vs Ambivert — clean lay-clinical framing for the swing axis.
• Cleveland Clinic — Otroverts: An Emerging Personality Type and Wikipedia — otherness construct background.
• Dr. Rami Kaminski via LADbible and the Otherness Institute's documented otrovert traits — source for the five otherness facets the quiz covers.

Items are clinical-source, not clinically validated. The disclaimer is on the results page for a reason.

License

Polyform Shield 1.0.0 — Copyright © 2026 Ashley Childress.

Personal, professional, and commercial use is permitted; monetization (sale, paid SaaS, rebranded resale) requires prior written permission.

Author

Ashley Childress — Senior Software Engineer @ The Home Depot. Builds odd, opinionated tools and writes about what AI is actually good for in production.

• GitHub: @anchildress1
• Site: anchildress1.dev

This quiz is a portfolio piece, not a product. If it broke, open an issue. If you want to talk about the architecture, the GHA wiring, or why the otherness axis is unipolar, reach out — those are the interesting parts.