A week to something you can click: what fits in it, measured on seven projects

Our approach page says a week to something you can click: not a deck, a working prototype in the real stack. That is a claim, and a claim without a number reads as marketing. So here is the number. Across seven projects started since October 2025, the first seven calendar days produced between 57 and 439 commits, between 13,616 and 104,874 lines, and between 7 and 41 screens a person could open. The same measurement on two projects we started in 2022 gives one commit and six.

What we measured

Seven repositories of ours, each from its first commit to the end of its seventh calendar day: a marketing site, a subscription product with a mobile app and a web admin, an internal inventory tool, a price comparison platform, a sports club site, a travel guide, and a one-to-one talk app. Four are clients', three are ours. All of them AI-native, all started between October 2025 and August 2026.

For each: commits and commit-days inside the week, lines added, the screens that existed at the last commit of day seven, and — the part that matters — the day on which each of the parts from the previous article first appeared: authentication, database migrations, tests, a CI pipeline, billing, email, translations. Then how much of the week's code is still in the repository today.

What a week is

• A marketing site — 141 commits on 5 days, 31,879 lines, 10 pages with a CMS behind them and 25 test files. Released on day three.
• The subscription product — 183 commits on all 7 days, 50,115 lines, 41 screens: an admin with a dashboard, programmes, meal plans, entitlements and notifications. Database on day two, login and a pipeline on day three.
• The inventory tool — 72 commits on 5 days, two people, 24,467 lines, 19 screens: inventory, people, reports, invitations, onboarding, transactions, organisation settings, login.
• The price comparison platform — 439 commits on all 7 days, one person, 104,874 lines, 29 screens, most of them admin: canonical products, categories, mappings, crawl, logs, matches. Login and database on day one.
• The sports club site — 69 commits on 3 days, 20,988 lines, 9 pages with a CMS.
• The travel guide — 57 commits on 4 days, 13,616 lines, 9 pages with a CMS.
• The talk app — 30 commits on 5 days, three people, 17,316 lines, 7 screens: two mobile tabs and an admin sign-in. The smallest week of the seven, and the only one started in 2025.

For contrast, the two 2022 projects: a booking platform's first week was one commit, followed by nothing for two months; a retail site's rewrite began with six commits and 99 lines. The approach page says the gap between an idea and a first version used to take a quarter and now takes weeks. Measured, the first week has grown by roughly two orders of magnitude.

What does not fit

Every one of the seven had screens with a real data model behind them at day seven. Here is what they did not have:

• Tests. Four of the seven had none at day seven. One got them on day 16. Three never have.
• A pipeline. Three of the seven had CI on day one. The others got it on days 2, 3, 9 and 10.
• Billing. In the price comparison platform — the 439-commit week — a payment provider's checkout and webhook were in place on day one. In the subscription product it arrived on day 28. In the other five, not at all.
• Translations. Day 3 in two projects, day 29 in another, day 114 in a fourth; never in three. In the inventory tool, adding a second language on day 29 moved every screen built in week one to a new path.
• Email. Day 1, day 1, day 3, day 7, day 197; never in two.

That is the boundary. A week buys the shape of the product: the screens, the data model, the admin, a deploy you can send a link to. It does not buy the parts that make it safe to charge for, which arrive over the following weeks and months on a schedule the previous article measured at seventeen percent of everything.

Week one survives

The instinct is that a prototype built this fast is thrown away. It is not. By exact file path, the share of each project's week-one files still in the repository today:

• Marketing site — 100%
• Travel guide — 100%
• Sports club site — 98%
• Price comparison platform — 94%
• Subscription product — 92%
• Inventory tool — 83%
• Talk app — 80%

Where a week-one screen is gone by path it has usually moved, not died: the inventory tool's screens went under a locale prefix, the talk app's under an authenticated group. Two real deaths in seven projects. The subscription product built meal plans in its first week and dropped them by a migration dated three weeks later. The price comparison platform built its data feed with one tool on day one and had replaced it with a small Rust program by day seven — a rewrite inside the week, not after it.

This is why the week has to be in the real stack. A prototype that survives at 80 to 100% is not a prototype; it is the first week of the product. Built in a throwaway tool, the same week would have been a deck with buttons, and week two would have started from zero.

What to do with the numbers

• Ask for the screens, not the plan. After a week there should be a URL, and a list of what is behind it. Seven projects say the list is between seven and forty-one items.
• Do not mistake the week for the product. At day seven, assume no tests, no pipeline, no billing, one language. Ask which of those is next and when. The honest answer is a date per part.
• Expect one rewrite inside the first month — a data pipeline, a screen structure, a feature that turns out not to belong. Two of seven had one. It is cheap at that point; it is the reason the week exists.
• Keep it. The code from week one is still running in every one of these. Treat it as production code from the first day, because that is what it becomes.