What we ship, decisions we make, questions we hear from homeowners, and Sander's in-the-wild domain observations.
Today appkeep opened for signups at €37 a year, and the waitlist closed; the people who waited were invited this morning with two months free. Before opening, we ran the whole money path ourselves on a real card: paid, took the receipt, then tested the exit by cancelling in one click, refunding, and deleting the account, and confirmed the card details were erased while the invoice is kept, as Finnish law requires. The full run caught one real gap on our own first invoice, before any customer paid: a missing Finnish tax registration that charged zero VAT. It is fixed. Three smaller improvements are filed, and a publishing job that had never run is running again.
We send one short email when a new guide goes up. For a stretch this month the link in three of them, all our Finnish guides, pointed at a page that was not there. We found it by reading our own send log, not from a complaint, fixed the address the emails are built from, and stopped the list from mailing a Finnish guide to someone who never asked for one. The deeper lesson is the one we are now building toward: a guide about one country should reach the people who live there.
The build is finished, so this week went into checking our own work before the doors open, the way we tell you to inspect a house before you sign. The pay path is whole and still switched off until we open: subscribe, receipt, and a one-click way back out. We closed the one outside call our own check found, a font service, so the site now reaches no one else while you read. We audited the code the way we would a house and closed the short list it returned the same day. And we wrote down, in plain words, how long we keep every kind of record. What is left is a date.
The payment-schedule guide is live: 30 percent maximum up front, tied to materials; the middle in tranches behind the work; at least ten percent held to the end. It comes from Sander's own payment memo, and nearly all of the memo shipped whole. The exception is the story: the memo's 'inspection runs eight to ten percent of job value' is true in his practice and would mislead in print, because the fee is flat and a percentage is not. We cut it, shipped the number inside a worked example instead, and asked him for the shape that would let it back in.
A number on its own is not an answer: 'is €640 fair?' is the actual question. This week the calculator learned to answer it: put in a quote, and you get a verdict against the fair band for that trade, not a bare number to squint at. The same verdict now reads back after the work is done, and when what you paid was sound, appkeep says so, because nobody else in this industry will ever tell you that you did well. Underneath, we verified the consumer statutes in Finland, Sweden and the Netherlands, so the price ceiling you send a contractor is written in your market's own legal vocabulary. And the calculator now emails you the mandate text, works in kronor, and hands off to the app for the rest.
You can own a house for twenty years and never hear, from anyone who knows, that a decision you made was sound. The contractor profits from the work, the insurer's silence only means no claim was filed, the neighbour is guessing; none of them has your life in mind, so the unease survives even the decisions you got right. This week we built the first surface that answers it. AppKeep already estimates what a job on your house should cost; now, when you record what you actually paid, it reads that back against its own estimate and tells you, in one plain line, whether it was in line. Under or in line: good. Over: context, never an alarm, because the money is already spent. We also wrote down, on the About page, why any of this exists.
We said appkeep opens to homeowners end of June. End of June came, and we did not open. This is the entry where we say that plainly: the date moved to mid-July, because the parts you would touch first were not yet things we would stake your trust on. What we built instead: the plan learned to bend: you can now move a job to the year you will actually do it, instead of arguing with a grid that thinks it knows better. And you can now see every assumption appkeep made about your house, and correct it. A tool that holds a record of your home has to take corrections without sulking. That felt more important than a date.
We open at the end of June, so this week went into the parts of the product you will never see and would only notice if they failed: the ones that decide whether we have earned the right to hold a record of your house. We built nightly encrypted backups that live off our main machine, then proved they work the only way that counts: we rebuilt the database from one, on a different computer, to be sure it comes back before we ever need it to. We tightened account creation so your record stays yours. And before writing a line of the paywall, we committed in writing to telling you before a renewal charges and making cancellation exactly as easy as joining. None of it shows from the outside, which is the point: the work that earns trust is the work nobody sees, done before you ask for it.
Two weeks from launch, we reviewed the product the way we tell homeowners to inspect a house, looking for where a claim and the thing it describes have drifted apart. We found two, the same shape. The waitlist page promised no drip campaign and one email, while the system was sending four useful guides over a signup's first two weeks; we kept the guides, fixed the promise, and added the unsubscribe link they should always have carried. And the mandate calculator offered a 'quick fix' price bracket that doesn't exist in real paid work, so we dropped it to the half-day floor. Closing the gap between what something claims and what it is is the product's whole job; this week we closed two of our own.
Two decisions, both about leaving things out. The plan screen now opens on the answer (what's coming and roughly when) with the controls a step back, so you reach for it and put it down instead of studying it. And distribution: we built the small tool that drafts a post when a guide ships, then chose not to automate posting to a company page with two followers, because the missing piece is an audience, not machinery. The same discipline we sell homeowners: fix the part that's failing, not the part that's old.
A week with no new features, spent making appkeep legible to machines. AI search couldn't find the site, and when it did, it confused appkeep with HomeKeep and a defunct Estonian app of the same name. The site turned out to be indexed but unranked, which is a different problem than not indexed, with a different fix. So the week went to identity: founder schema, a LinkedIn company page, inbound links, and verification in both Bing and Google, none of which shows on the site.
Built a mechanical audit for the writing this week, ran it across the twenty guides, and found that the AI tell manual review misses most often is not em-dashes. It is parallel sentence openings, three or four in a row starting with the same word. Then Sander reviewed one of the guides and caught three patterns the script will never reach.
appkeep's plan projects replacement costs across seventy-eight component variants, each with a lifespan range attached. For months those numbers carried a flag that said 'Sander to validate.' This week we noticed that was the wrong ask, and rebuilt the validation process around who actually sources lifespan data and who signs off on it.
DMARC, Google Search Console, Open Graph, a byline on every guide, and ten other plumbing jobs that together turn a newly launched site from 'exists' to 'findable.' The work nobody pays attention to until it's missing.