Experiment Decision Log

Registro de decisão para experimentos: hipótese → evidência → decisão → aprendizado.

Decision log for product experiments: hypothesis → evidence → decision → learning.

PT-BR · English · Live Demo · Stack · Architecture · Quick Start · Author

Live Demo · Repo · Portfolio · LinkedIn

Lab / demo notice: frontend-first A/B decision lab with localStorage. Approximate proportion tests with caveats — not a production experiment platform, feature flagger or formal statistical design tool.

---

PT-BR

Visão geral

O Experiment Decision Log transforma testes A/B em decisões documentadas: hipótese, métricas, guardrails, evidência (com caveats), decisão humana, timeline e relatório Markdown.

Problema

Muitos projetos param no p-value. Significância estatística não é decisão — falta registrar risco, guardrails, custo e aprendizado.

Para quem

Product analysts, growth/PMs e data scientists que precisam fechar o ciclo de experimento com responsabilidade.

Funcionalidades

• Cadastro de hipótese, variantes, métrica primária e guardrails
• Análise de proporções (uplift, IC aproximado, nível de evidência) com linguagem cautelosa
• Framework de decisão (rationale, impacto em guardrails, riscos, follow-up)
• Timeline e export Markdown
• 4 demos: ship forte, inconclusivo, trade-off conversão vs AOV, amostra insuficiente
• Tour guiado e case sintético
• Testes Vitest das regras estatísticas e de decisão

Escopo e limites (honestos)

• Sem backend multi-usuário, sem tracking/split real
• Estatística simplificada para proporções (não substitui design formal de alto risco)
• Guardrails ainda qualitativos no MVP
• Persistência em localStorage

---

English

Overview

Experiment Decision Log turns A/B tests into documented decisions: hypothesis, metrics, guardrails, evidence (with caveats), human decision, timeline and Markdown report.

Problem

Many projects stop at the p-value. Statistical significance is not a decision — risk, guardrails, cost and learning still need a record.

Who it is for

Product analysts, growth/PMs and data scientists who need a responsible experiment close-out loop.

Features

• Hypothesis, variants, primary metric and guardrails
• Proportion analysis (uplift, approximate CI, evidence level) with cautious language
• Decision framework (rationale, guardrail impact, risks, follow-up)
• Timeline and Markdown export
• Four demos: strong ship, inconclusive, conversion vs AOV trade-off, underpowered sample
• Guided tour and synthetic case
• Vitest coverage for statistics and decision rules

Scope and honest limits

• No multi-user backend, no real tracking/split
• Simplified proportion statistics (not a substitute for high-stakes formal design)
• Guardrails are still qualitative in the MVP
• Persistence is localStorage

---

Live Demo

Surface	URL
Public lab	https://experiment-decision-log.vercel.app
Guided tour	/tour
Promo case	/cases/promo-aov
GitHub	https://github.com/BarujaFe1/experiment-decision-log

How to try: open the lab → load demos → open analysis (note caveats) → fill the decision panel → export Markdown. Use Carregar dados demo to reset localStorage.

---

Screenshots

Home	Experiments
Analysis	Decision panel
Methodology	Tour

_{Promo / AOV case}

---

Stack

Layer	Technology
App	Next.js 15, React 19, TypeScript, Tailwind CSS 4
Domain	Zod, custom statistics + decision rules, localStorage
Charts / tests	Recharts, Lucide, Vitest, Testing Library

---

Architecture

src/
  app/            pages (home, experiments, cases, tour, methodology)
  components/     decisions, experiments, metrics, layout
  lib/            experiment-model, statistics, decision-rules, storage, markdown-export
  tests/          Vitest
assets/           icon, hero, screenshots

Flow: hypothesis → results → evidence classification → human decision → timeline → Markdown report.

---

Quick Start

Prerequisites: Node.js 20+, npm, Git.

git clone https://github.com/BarujaFe1/experiment-decision-log.git
cd experiment-decision-log
npm install
npm run dev

Open http://localhost:3000.

npm test
npm run typecheck

---

Technical decisions

• Decision over calculator: auto recommendation is a suggestion; UI requires human rationale
• Cautious statistics language with caveats for underpowered / zero-division cases
• Demo narratives that include trade-offs and inconclusive outcomes (not only “wins”)
• Frontend-first storage ready to evolve to API/SQLite later

---

Roadmap

• Backend SQLite / FastAPI
• Continuous metrics and revenue-per-visitor
• CSV import + pre-test templates
• Simple power analysis
• Light auth for teams

---

Author

Felipe Alirio Baruja — data / product / full-stack portfolio.

• Portfolio: https://barujafe.vercel.app/
• GitHub: https://github.com/BarujaFe1
• LinkedIn: https://www.linkedin.com/in/barujafe/

---

License

MIT — see LICENSE.