What I've been building

Today was about making a finished habit feel good, and reaching the ones that had gone missing.

• The little celebration when you finish a habit now plays on the row you tapped, so you see it instead of missing it off the bottom of the screen.
• A progress line at the top of Today shows how much is left, and clearing the last habit ends the day with a small finish.
• You can now scroll to and log older habits in the “Log another habit” list, which used to sit off the bottom of the screen with no way to reach them.

Rules that used to live only in documents became automated checks that fail a bad change on their own.

• A worker’s deploy settings are now checked against its code, so a setting missing from the code fails the build, not at runtime.
• A worker’s compressed size is checked against a budget, so a heavy dependency can’t quietly bloat what gets deployed.
• Unused code and dependencies are now flagged, so leftovers from a refactor don’t pile up unnoticed.
• The document listing a worker’s routes and connections is generated from the code, so it can’t quietly fall out of date.
• A message or webhook can now be proven to run once, even when two arrive together, using a shared check any project can reuse.
• Each screen is checked against a saved image, so an accidental layout shift is caught before it reaches anyone.
• Each screen is scanned for accessibility problems that block real users, the first such checking across any of the projects.

Cloudflare caps how many external calls one request can make, and going over crashes it only once real traffic arrives.

• A test can now set a call budget and fail the moment a request goes over it, so the problem surfaces while you build.
• Fixed the front-end build, which broke when a shared testing package started pulling in server-only code.

Golden gained mutation testing, which breaks the code on purpose and checks that a test notices the change.

• Running it on golden’s own code shows which tests would catch a real bug and which pass no matter what, so weak tests stop hiding.
• Any new app built from golden starts with the same setup ready, so its tests can be judged the same way from day one.

Anecdote texts a reminder when a saved story still needs its result or its next action written down.

• You can now tap that reminder to open the exact story, instead of opening the app and searching for the right entry.
• Fixed the reminder still telling you to reply to archive a story, when replies to that number no longer reach anyone.

The work built a complete example app whose only job is to be the pattern every new project here is measured against.

• The example covers a real app end to end: creating and editing notes, a daily stats view, sign-in, and a paid AI call.
• A new build check fails when an app’s reply stops matching its agreed shape, so the mistake is caught before release, not by a user.
• The project gained its first automated browser test, which drives the whole running app to catch breaks that smaller tests miss.

The plan for golden was written and refined, a single working example app that new projects copy from instead of repeating setup and drifting.

A code review checked how the planner held up after its recent features, and mapped out the fixes and tidy-up that come next.

This was groundwork rather than a visible feature, covering the text-message conversations, scheduled jobs and account checks that make up Anecdote.

The toolkit that measures how the apps behave now follows reads and writes to the cache, the fast store that avoids repeat work.

Behind-the-scenes monitoring now records more detail about how the app’s parts work together, so faults can be traced and fixed faster.

The backend behind the site gained detailed timing records, so a slow response can now be pinned to the step that caused it.

The service that routes text messages and AI calls now records how long its internal steps take, so a slow request can be traced.

Behind the sign-in flow, the service now records its own activity in more detail, naming the Google token step and grouping requests by route rather than by user id.

Storymap now records its own internal steps in more detail, so a slow or failing request can be traced to the operation behind it.

The shared tracing tools that measure the live apps gained a more detailed view of where each request’s time goes.

• Traces now group requests by their route pattern instead of the exact web address, so every hit to one screen lines up together.
• You can now see the time spent handing a request from one service to another, which was hidden before inside the receiving service.
• Database traces now show how many rows a query read, so one that scans a whole table stands out instead of looking fast.

Storymap got a nudge to where it opens and a fix that protects work you haven’t saved yet.

• Opening storymap now takes you straight to the Planner board, instead of the general workspace board it showed before.
• Saving no longer fails when you close an item then reopen it before saving, which used to leave your edits stuck and unrecoverable.