Special case

Product Studio

I work alone, on large requests and short deadlines.

The product team I built with AI: twelve specialist agents in Claude Code. My bet: structured AI can extend what one person working alone delivers, without replacing human judgment. The system stops when the decision is mine and only counts a delivery as done with proof.

See the deliveries

  • 24days

    for a full product, from discovery to publication: Nexo Quality, a concept project.

  • 4days

    from request to presenting a client’s new website as a navigable prototype.

  • 12agents

    specialists, from research to audit. Ten have already worked on real projects.

  • 83days, 10 versions

    to build the system through version 1.0, counted in calendar time, not hours of work.

Role

Product Designer. Design and build.

Use

My own. Not publicly available.

Built on

Claude Code, by Anthropic.

Version

v1.0, October 2026

Limits: time saved has not been measured, and two of the twelve roles have not worked on real projects yet.

Highlights

Why it exists

The problem

Design, development and QA all land on me. AI helps, but without control it fails in expensive ways.

In testing, it spent past the ceiling, skipped mandatory steps and called work done without proof.

System principle

Existence is not evidence. Completion requires proof.

The bet

Sizing the process to each request and demanding proof for every delivery can help someone working alone meet short deadlines.

In real use

Three scales: a full product, a client site and the requests of this portfolio. The first two ran on versions before 1.0.

Deliveries

A product, a client site and this portfolio

  1. Full product

    Nexo Quality

    September 4 to 27, 2026 · 24 days

    A concept project: an automotive quality platform with a fictional company and data. The agents worked from discovery to final QA, and I decided at every phase. The second round of the low-fidelity prototype came from Codex, another AI tool.

    See the caseOpen the product

    • Product definition audited by someone who did not write it
    • Final QA in three browser engines
    See the demo
    Interactive demo: the product screens, in HTML. Open a case to see its Summary and Timeline.
    Skip the demo
    Search Nexo…CtrlK

    Cases

    Default sort: Priority
    Investigationno active flagsPriority:HighP. Rocha (SQE) · Plant SP1
    TriageCriticalPriority:HighJ. Souza (SQE) · Plant SP1
    Waiting on supplierBlockedPriority:HighP. Rocha (SQE) · Plant RS1
    Internal analysisCriticalEscalatedPriority:HighJ. Souza (SQE) · Plant SP1
    Waiting on supplierno active flagsPriority:HighP. Rocha (SQE) · Plant RS1
    Containmentno active flagsPriority:MediumP. Rocha (SQE) · Plant MG1
    InvestigationOverduePriority:MediumJ. Souza (SQE) · Plant SP2
    Implementationno active flagsPriority:MediumP. Rocha (SQE) · Plant SP2
    Waiting on supplierno active flagsPriority:MediumJ. Souza (SQE) · Plant SP1
    Verificationno active flagsPriority:LowJ. Souza (SQE) · Plant MG1
    Closedno active flagsPriority:LowP. Rocha (SQE) · Plant SP1
    NewIncomplete dataPriority:—— unassigned — · Plant SP1

    NC-000123 — Clearance out of specification — suspension bushing

    TriageCriticalPlant SP1Owner: J. Souza (SQE)Priority:High

    Summary: Initial data recap

    Material/Part
    PN-55019-B · Rear suspension bushing (Rev. B)
    Lot
    L-FORN-2290 / L-INT-5541
    Supplier
    AutoPeças Zenith S.A., Sorocaba/SP
    Material criticality
    High
    Initial segregation
    Yes, 2026-09-12 08:40, by R. Nunes

    Severity

    Proposed severity (Inspector)
    High
    Confirmed severity (SQE)
    Critical
    Recorded justification
    Potential suspension safety impact — raised to Critical by J. Souza.

    Critical severity: the Critical flag becomes visible to the Coordinator to confirm/act (BR-14).

    Problem Solving scope

    Priority

    Proposed priority
    High
    Why?

    Factors: material criticality (High) + proposed severity (High) + lot volume.

    Override

    Outside this role's approval authority — priority override (SQE/Coordinator).

    — no override —

    Actions — Waiting on you

    A crop of the product's Actions screen, as SQE P. Rocha sees it.

    NC-000890·Corrective action·Submit Corrective action evidence

    Incorrect engine mount fastening — in implementation

    Owner: SQE P. RochaDue date: 2026-09-14Task status: In progressNo blocker

  2. Client

    3 Lírios website

    August 27 to 31, 2026 · 4 days to the presentation

    A high-fidelity navigable prototype of the new website of a housing builder. Presented in 4 days and approved by 3 Lírios after 4 weeks of adjustments. Version 1 is live.

    See the caseOpen the site

    • One internal review, disclosed at the decision point
    • A scrolling library caused defects and was swapped out the same day
    See the demo

    On a phone, this demo scrolls inside itself. To keep going down the page, swipe on the area outside it.

    Skip the demo

    Interactive demo: the site in HTML, in Portuguese, with its own scrolling. Filter the homes, open a listing, run the pre-assessment.

    View on

    Parceiro
    Imagem ilustrativa

    Sair do aluguel · Investir · Imóveis financiáveis

    Sua casa própria
    começa aqui.

    Imóveis para diferentes momentos e objetivos: seja para sair do aluguel, conquistar sua casa própria ou investir. Ajudamos você a entender as possibilidades de financiamento e encontrar uma opção que faça sentido para o seu momento e orçamento.

    Gratuito, sem compromisso e sem consulta ao seu CPF.

    Como funciona

    Um caminho previsível até a primeira casa.

    1. Você faz uma pré-análise rápidaPoucas perguntas para entendermos seu momento e seu perfil de renda.
    2. Ajudamos você a encontrar o imóvel ideal para o seu perfilVocê recebe as opções que combinam com o que precisa e com o que cabe no orçamento.
    3. Você entende o financiamento MCMV e demais linhas de crédito disponíveisExplicamos entrada, uso do FGTS e os próximos passos, sem burocracia.

    Imóveis à venda em destaque

    Três começos possíveis, com o preço na frente.

    Situação da obraNa plantaEm obraPronta para morar
    Casa Orquídea, casa térrea de 2 quartos na Vila Flor.
    Pronta para morarImagem ilustrativa
    Casa Flora, casa térrea de 2 quartos em Nova Pescaria.
    Em obraImagem ilustrativa
    Casa Peônia, casa térrea de 2 quartos na Vila dos Araçás.
    Em obraImagem ilustrativa

    Financiamento Minha Casa Minha Vida

    Feito para caber no seu orçamento.

    Três coisas tornam a compra possível — e todas cabem no Minha Casa Minha Vida.

    Feito para o Minha Casa Minha Vida

    O programa cobre a maior parte do valor; você entra com uma fração.

    Use seu FGTS

    O saldo do seu FGTS pode compor a entrada ou abater parcelas.

    Entrada facilitada

    Condições de entrada parceladas, combinadas conforme o seu perfil.

    Feito para o Minha Casa Minha Vida

    O programa cobre a maior parte do valor; você entra com uma fração.

    Use seu FGTS

    O saldo do seu FGTS pode compor a entrada ou abater parcelas.

    Entrada facilitada

    Condições de entrada parceladas, combinadas conforme o seu perfil.

    Imóveis entregues

    Seu lar, sua história.

    Fachada térrea da Casa Camélia, com revestimento de pedra clara e porta de madeira.
    Fachada térrea branca da Casa Melissa, com detalhe em cinza.
    Fachada da Casa Rosa, branca, com platibanda cinza e porta de madeira.

    Sobre nós

    Realizando o sonho da casa própria

    Uma empresa que nasce com o objetivo de realizar o grande sonho da casa própria. Queremos que você também tenha sua casa para viver feliz e com muita saúde.

    Cuidamos da qualidade de cada detalhe do seu patrimônio, sempre com atenção, responsabilidade e compromisso. Nossa equipe tem grande experiência no mercado e está ao seu lado do início ao fim. Acompanhamos cada etapa do processo, oferecendo atendimento próximo, orientação e suporte para que você tenha segurança e tranquilidade desde o primeiro contato até a entrega da sua casa.

    Nosso compromisso é entregar no prazo, cuidar de cada detalhe e fazer com que a realização do seu sonho seja uma experiência segura e tranquila.

    Casa Camélia, entregue pela 3 Lírios, com fachada clara, porta de madeira e céu azul.
    Casa Camélia, entregue

    Depoimentos

    Quem já mora, recomenda.

    Histórias de quem realizou o sonho da casa própria com a 3 Lírios.

    Eu achava que financiar era coisa complicada. A equipe explicou cada etapa e, em poucos meses, eu estava com a chave na mão.
    Mariana S.Tubarão — SC
    Fizemos a pré-análise em minutos e já sabíamos até onde podíamos ir. Sem surpresa no meio do caminho.
    Rafael e JulianaCapivari de Baixo — SC
    A casa ficou do jeito que mostraram. Entrega no prazo e todo mundo atendeu com paciência.
    Dona NeusaLaguna — SC
    Usei o FGTS na entrada e nem sabia que podia. Me explicaram tudo no WhatsApp, sem enrolação.
    Carlos H.Tubarão — SC
    Meus filhos ganharam um quintal. Era o sonho que a gente adiava há anos.
    Patrícia M.Laguna — SC
    Atendimento do começo ao fim. Sempre que eu tinha dúvida, alguém respondia rápido.
    Anderson L.Capivari de Baixo — SC
    Saímos do aluguel com a parcela cabendo no orçamento. Valeu cada visita à obra.
    Fernanda e BetoTubarão — SC
    Casa simples, bem-feita e com tudo no lugar. Recomendo para quem quer sossego.
    Seu OsmarLaguna — SC

    Depoimentos de demonstração.

    Corretor parceiro analisando a planta de um empreendimento sobre a mesa.Imagem ilustrativa

    Escritório Virtual do Corretor

    Tudo o que o corretor parceiro precisa, num lugar só.

    Ferramentas, materiais e acompanhamento para você levar o cliente da primeira simulação até a entrega das chaves — sem burocracia.

    Pré-análise na hora

    Rode a simulação do cliente e mostre o valor financiável na mesma conversa.

    Materiais de venda prontos

    Plantas, tabelas de preço e peças de divulgação sempre atualizadas para enviar.

    Acompanhe suas vendas

    Propostas, reservas e comissões num painel único, do lançamento à entrega.

    Time da 3 Lírios com você

    Suporte direto em cada etapa do fechamento, com resposta rápida.

    Fale com a gente

    Prefere conversar agora? Chame no WhatsApp.

    3 Lírios Construções e Incorporações

    Sua casa própria pode começar aqui.

    Navegação

    Contato

    • 3liriosconstrucoes@gmail.com

    3 Lírios Construções e Incorporações · CNPJ 00.000.000/0001-00 · Rua Exemplo, 000 — Palhoça/SC (dados de demonstração)

  3. Day to day

    This portfolio

    September 29 to October 5, 2026 · 6 days

    Fourteen requests with a record: cases, color, background, navigation, loading screen and this page.

    See the record of one request

    • In 4, I rejected or sent back what the system delivered
    • In 4, QA or the independent critique found defects before publishing

Record of one request

The "Cases" item on this site, from request to publishing

A small request from October 4, 2026 shows the whole process: the Cases item in this site’s navigation. Pick a step to see its real record.

  1. Step 01 of 10

    WhoMe

    I asked whether a "Cases" item in the navigation was worth it, with a panel of my cases, and pointed to a reference site.

  2. Step 02 of 10

    WhoExecutive Orchestrator

    A medium task on an existing product. It chose the Lean path, planned four specialists and set a mandatory visual stop: the reference was shaping the direction.

    On record

    • Path: Lean
    • Specialists planned: 4
    • Specialist ceiling: 220,000 tokens

  3. Step 03 of 10

    WhoSystem, with an automated browser

    Before offering an opinion, it opened the reference site and measured the panel, which opens on hover and becomes a text list on mobile.

    On record

    • Panel: 682 × 368 px
    • List: 322 px, rows of 58 px
    • Preview: 360 px

  4. Step 04 of 10

    WhoUX Architecture Lead

    Recommended with adjustments and recorded the argument against: with three cases the gain is small, because the Home already shows them.

    On record

    • Opinion: recommended with adjustments
    • On a phone: text list, no image

  5. Step 05 of 10

    WhoUI Lead and Interaction Lead, in parallel

    One specified the panel, the other the behavior with mouse, keyboard, touch and reduced motion. The two diverged on an 8 px gap, resolved without changing the look.

    On record

    • Panel: 760 × 436 px
    • A 340 px list and a 420 px preview
    • 26 acceptance criteria

  6. Step 06 of 10Waiting for a decision

    WhoMe

    The system stopped before building and asked two questions: who builds it, and what a click on the item does. I answered "use the method you think is best" and "it only opens the panel".

  7. Step 07 of 10

    WhoSystem, no specialist

    It built the panel and measured the result in the browser, at three widths. It tested mouse, keyboard and touch and checked contrast with a script.

    On record

    • Panel measured: 760 × 436 px
    • 11 color pairs, the lowest at 6.28:1
    • No text under 16 px

  8. Step 08 of 10Waiting for a decision

    WhoMe

    I approved it with changes. The subtitles did not stand out and the titles were too heavy. The system rewrote the three subtitles and changed the title font.

  9. Step 09 of 10

    WhoQA and Release Lead

    It redid the critical tests on its own. Approved with conditions, no serious defect, and declared what it did not test: touch on a real device, screen reader and other browsers. Cost passed the alert line and was recorded.

    On record

    • Verdict: passed with conditions
    • 1 minor finding, no fix needed
    • Cost: 190,021 of 220,000 tokens

  10. Step 10 of 10Waiting for a decision

    WhoMe

    Nothing goes live without an explicit request. I asked for one last change to the menu order and wrote "you can publish". It was published and checked in production the same day.

    On record

    • 10 pages checked in production
    • Total cost: 278,113 tokens, with classification

On the yellow steps, the system stopped and waited for my decision.

Cumulative cost of specialists

190,021 of 220,000 tokens

Tokens are the unit of AI usage. The ceiling is set before the work starts.

From request to delivery, at the right size

I write the request as I would for a team. An orchestrator agent always follows the same six steps and calls in only the specialists the request needs. The decisions listed at the top are mine; the system recognizes them and asks.

  1. Understands the request

    Checks sources and marks what it does not know.

  2. Picks the path

    Direct, Lean or Full, by size.

  3. Calls in specialists

    Only the ones needed, with minimal context.

  4. Measures cost

    Checks the budget before every call.

  5. Stops and asks

    Scope, visual direction, budget and publishing wait for my answer.

  6. Proves and delivers

    QA verifies with evidence; only then does the delivery count.

Four rules that hold the system together

  1. Human decision

    Rule

    The system recognizes the decision point, stops and presents the options. It never approves on its own.

    Evidence

    In the final test, 2 pauses for budget waited for real answers before work continued.

  2. Selective context

    Rule

    Each specialist gets only what the decision requires and records the reason if it reads more. It is not reliable yet: in tests, specialists read more than needed.

    Evidence

    Average cost per call had risen 79% when each specialist read the full instructions.

  3. Cost Guard

    Rule

    Before each call, it checks the budget; afterward, it records the real cost. At the ceiling, it stops and waits for a human decision.

    Evidence

    One cost test reached 117,244 tokens, 27,244 above its 90,000 ceiling.

  4. Completion by evidence

    Rule

    Done requires three things together: the output, the verification and the evidence left behind.

    Evidence

    The auditor rejected the release candidate 3 times; the last one only for lack of a real test.

My role, decisions and limits

My role

I designed and built the system for my own work. In use, scope, direction and publishing are my decisions.

Decisions I made

  • Instead of closing the release with the final test pending, I required a real test in a fresh session. Criterion: done requires proof. Cost: the release waited for that test and for two budget increases. Consequence: the auditor rejected the candidate 3 times, the last one only for lack of that test.
  • I rejected the visual direction of one test: the process had passed, but the visuals had not. Criterion: what you see, not what the process completed. Cost: about 550k tokens already spent before the first honest visual critique. Consequence: the independent visual critique became mandatory.
  • I authorized, one by one, the two budget increases of the final test, instead of letting the system go past the ceiling. Criterion: the ceiling is the rule and only I can release an exception. Cost: less autonomy. Consequence: both pauses waited for my answer.

Known limits

  • The full product and the client site ran on versions before 1.0, without its cost record. The 83 days are calendar time, not hours.
  • In real use, the Lean path went over its ceiling more than once and needed my release. The final test was not blind, and formal validation is small.

The system up close

Requests, build history, routing, agents, toolkit and next steps. Open only what you want to check.

What was asked and what happened

Fourteen requests, each with what I asked and what happened. Ten are live.

Full square: it is live. Empty: it was not published.

  1. Sep 29

    Review the copy of the cases

    Published the same day.

  2. Sep 29

    Color: yellow and color photos

    The independent critique flagged 2 problems that did not exist. They were checked and dismissed.

  3. Sep 30

    3 Lírios case

    The independent critique scored it 27 of 40 and flagged 3 important problems. After the revision, 30.5 of 40.

  4. Sep 30

    Raise the level of the other cases

    QA passed it and flagged 1 minor defect, fixed before publishing.

  5. Sep 30

    Nexo Quality: explanation on cards and sidebar

    QA found 1 layout defect. It was fixed, and only then published in Portuguese and English.

  6. Oct 2

    Background with gradient and animated lines

    I approved the direction for the case pages.

  7. Oct 2

    More artistic lines in the background

    I rejected both variants. Neither was published.

  8. Oct 2

    Direct interaction in the case frames

    QA rejected the first delivery: on a phone, a map trapped the page scroll. Fixed and rechecked.

  9. Oct 3

    Loading screen with the logo

    I rejected two versions before the one that is live.

  10. Oct 4

    "Cases" item in the navigation

    I approved it with changes. It is the record shown above.

  11. Oct 5

    This page, first version

    I asked for a full review: I was not sure it explained the whole system.

  12. Oct 5

    This page, second version

    I asked for it to be completed: the technologies, the libraries and everything the system has were missing.

  13. Oct 5

    This page, third version

    I asked whether it would make a recruiter’s eyes light up. The assessment was not yet, with four problems named, and I approved the changes.

  14. Oct 5

    This page, current version

    It is what you are reading.

83 days, 10 versions

From the first file, on July 15, 2026, to version 1.0, on October 5. Each version records the failure behind it.

  1. Jul 15 to 24

    Architecture and roles

    Twelve roles with written contracts, activated, and a single command that receives the requests.

  2. Aug 3 to 10

    Visual quality and cost

    Versions 0.9.2 to 0.9.4: cost control and the rule of seeing pixels before deciding.

  3. Aug 26

    Proof and proportion

    Versions 0.9.5 and 0.9.8: completion by evidence, selective context and the three paths.

  4. Oct 5

    Version 1.0

    The candidate was rejected by the auditor. Fixed, tested and frozen.

Two failures that changed the architecture

Each row is a measured failure and the change it caused.

  1. The process passed, the visuals failed

    ~550k

    tokens spent before the first honest visual critique

    Change

    See before deciding. No visual direction moves forward without real pixels of the concept and a human decision.

  2. The release candidate was rejected by the auditor

    5.46x the ceiling

    1,201,867 tokens against a ceiling of 220k

    10 specialist calls ran after a pause that was mandatory.

    Change

    A real Cost Guard. Whoever calls the specialists measures real cost at every step and stops at the ceiling.

Three paths, one for each size of request

Every request is classified before it runs. Lean is the default; Full only comes in when risk or complexity call for it.

Request

Classification

  • Where the work happens: design, code or both
  • Expected level of finish
  • What is fixed, negotiable or open

Direct

A contained task, checkable on the spot, with no specialist.

  • Specialists: 0
  • QA: simple check
  • Independent audit: no
  • State summary: no

Lean

The default: a few specialists and limited risk.

  • Specialists: 1 to 3, usually
  • QA: proportional
  • Independent audit: light, if risk calls for it
  • State summary: if it spans sessions

Full

Only when risk or complexity call for broad coordination.

  • Specialists: 6 of 12 in the test
  • QA: formal
  • Independent audit: according to risk
  • State summary: required across sessions

Delivery with evidence

completion by evidence

or

Budget pause

waiting for a human decision

Cost Guard. Only on paths that call in specialists.

Full square: always. Half full: under a condition. Empty: no. Line weight shows how broad the process is, not its cost.

Twelve specialist agents, by discipline

Each role is a Claude Code subagent with a written contract: what it delivers, what it does not decide and when it must stop. Only the ones the request needs are called in.

10/12

roles have already worked on real projects. Missing: back end, and data analysis, which only comes in with a product in production.

Each segment is a role, in the order of the list below. Full: already worked on a real project. Empty: not yet.

  1. Product and research

    • Executive Orchestrator

      Sizes the request and picks the path.

    • Product Lead

      Scope, requirements and success criteria.

    • Research and Strategy Lead

      Gathers facts and marks what is missing.

    • Analytics and Growth Lead

      Only with a product in production and real data.

  2. UX

    • UX Architecture Lead

      Information architecture, flows and wireframes.

  3. UI and interaction

    • UI and Design Systems Lead

      Visual direction, components and tokens.

    • Interaction and Prototyping Lead

      States, keyboard, touch and motion.

  4. Development

    • Frontend Engineering Lead

      Turns what is approved into code.

    • Backend and Platform Lead

      Architecture, APIs, data and integrations.

  5. QA, documentation and audit

    • QA and Release Lead

      Tests and gives the verdict. Does not fix what it evaluates.

    • Knowledge and Documentation Steward

      Keeps the state summary and the sources.

    • Independent Auditor

      Checks the final delivery. Read only.

What it knows how to apply

The toolkit has 24 skills and, per discipline, methods, technologies and deliverables. Open it to see what has been used on real work.

61/76

toolkit items have already been used on real work. The rest are ready and have no recorded use yet.

Full square: already used on real work. Empty: in the toolkit, no recorded use yet.

  1. Product and research

    7 of 11 already used

    Technologies and tools

    • Four research specialists on demand: business, market, competition and behavior
    Methods and deliveries

    Methods

    • Evidence and assumption registers
    • Versioned product definition
    • Permission matrix by role
    • User validation plan, prepared and not run
    • RICE prioritization and INVEST stories
    • Four-voice council decisions
    • User research

    Deliveries

    • Problem framing
    • Product definition
    • Fact bank with a source on every line

  2. UX

    10 of 10 already used

    Technologies and tools

    • Navigable HTML prototype
    Methods and deliveries

    Methods

    • Information architecture and object model
    • Flows and low-fidelity wireframes
    • Evaluation by four methods: cognitive walkthrough, heuristics, domain audit and adversarial test
    • Operational scenarios by role
    • Interface writing
    • A check of every claim against the source

    Deliveries

    • Information map
    • Flows and wireframes
    • Findings report by severity

  3. UI and design system

    12 of 16 already used

    Technologies and tools

    • shadcn/ui
    • Tailwind CSS
    • Radix and Base UI
    • shadcn component registry
    • 21st.dev
    Methods and deliveries

    Methods

    • Three concepts rendered before choosing a direction
    • Independent visual critique by two isolated reviewers
    • Measurement of real references in the browser
    • Scripted contrast checks
    • Component coverage map
    • Adopt, adapt or create decisions for components
    • Style and palette library
    • Theme creation

    Deliveries

    • Art direction brief
    • Design system foundations
    • Interface specification

  4. Interaction and motion

    8 of 10 already used

    Technologies and tools

    • CSS and View Transitions
    • GSAP
    • Lenis
    • anime.js
    • motion/react
    Methods and deliveries

    Methods

    • States, keyboard, touch and reduced motion
    • Duration and easing tokens
    • Complex choreography

    Deliveries

    • Behavior specification
    • Loading animation

  5. Front end and platform

    11 of 15 already used

    Technologies and tools

    • React 19
    • Next.js
    • Astro
    • TypeScript
    • Vite
    • Tailwind CSS 4
    Methods and deliveries

    Methods

    • A representative slice before the full page
    • Small, isolated changes, measured in the browser
    • API contracts
    • Back-end patterns
    • Security review
    • Test-driven development

    Deliveries

    • Web application
    • Website
    • High-fidelity navigable prototype

  6. QA, documentation and audit

    13 of 14 already used

    Technologies and tools

    • Automated browser (Playwright, also as an MCP server)
    • Contrast script
    • DevTools inspection
    • axe, for accessibility rules
    Methods and deliveries

    Methods

    • QA in three browser engines
    • End-to-end test by user role
    • Reflow at 320 px, touch targets, focus order and reduced motion
    • Independent, read-only audit
    • Decision log and project state summary

    Deliveries

    • QA report with a verdict
    • Audit with findings by severity
    • Frozen version with a checksum
    • User manual in PDF
    • Documents in .docx and .pptx

Working environment

Connections from my account, outside the system: Vercel, to publish with a rollback point, and Figma through MCP, used in two case reviews. They are not Product Studio capabilities.

Next steps

  1. Recalibrate the Lean path budget

    Use real costs so it stops pausing before it builds.

  2. Measure the Full path

    The only formal data point predates the v1.0 rules.

Another case

Masterplan Urubici

Navigation by theme, not by department, and the same structure in 3 towns.

Role: Product Designer (end-to-end)

Web · Mobile · 0 to 1 · 3 towns

Masterplan Urubici portal home page, with the headline and the town's vision for the future