focusstationShippedFocusStation v0.1.1 — Shipping Reveals Everything You Missed
Historical note: this post records the reliability pass after v0.1.0. The completed app later changed sleep handling again: timers now continue accurately through screen sleep and lid-close, while the local midnight boundary still ends the day's session. It also gained daily workspaces, carry/repeat, history export, and a rebuilt UI architecture. Read the v1.0.0 closing build log →
v0.1.0 shipped. Then I actually used it.
I was proud of v0.1.0. A menu bar timer built over a few weekends by someone who'd never written Swift. Nineteen files, zero dependencies, shipped.
Then Monday came. I sat down to work and the timer froze. The popover stopped updating. The menu bar went blank. Tasks kept reordering themselves whenever I started one. Editing a running task was impossible — clicking the text field did nothing. Closing my laptop mid-timer made the clock stop.
Every single bug was something I missed because I was testing in isolation. Start a timer, verify it runs, move on. I wasn't living with the app. Real usage exposes everything.
v0.1.1 is five days of fixing what v0.1.0 got wrong. Same feature set. Everything actually works now.
What was broken
The sleep timer
Close your laptop mid-timer. Open it three hours later. v0.1.0 showed the same elapsed time as when you closed it. Your 2-hour meeting still said 2 hours.
The code was actively pausing timers on sleep. For v0.1.1, I treated that as a timer-engine bug and let Date() advance through sleep. Thirteen lines deleted. Zero lines added.
Real use challenged that decision again. Pausing on sleep meant turning off the screen silently changed task state and forced a manual Resume click after wake. The completed app returns to the timestamp model: sleep does not mutate a running task, elapsed time catches up after wake, and the local midnight boundary still caps the session to its scheduled day.
The frozen popover
Elapsed time in the dropdown only updated when you moved the mouse or clicked something. Not smoothly. Not every second.
The app was destroying and recreating the entire popover every single second. Every timer tick nuked your edit state, your hover state, your scroll position. Fix: update the label only. Let the popover live. Six rewrites to get right.
The blank menu bar
Start a task. Popover says it's running. Menu bar shows just a brain icon.
The status item width wasn't updating when content changed. Fix: measure the content, set the width manually, clamp between 24-260px. Called on every tick and every state change.
Other things that broke
- Editing a running task did nothing — background state sync was clobbering the edit form. Fixed by pausing sync during edits.
- Starting a task jumped it to the top of the list — sort was by activity, not insertion order. Fixed by sorting on insertion order only. Drag-and-drop is now the only way to reorder.
- No visual signal when you blow past your target — set 30 minutes, hit 2 hours, timer color never changed. Now elapsed time turns red when you're past target.
What I learned
"It works on my machine" is a lie. Every bug was something I missed testing in short bursts. I wasn't sleeping my Mac mid-timer. I wasn't editing tasks while one was running. I wasn't letting the popover stay open for an hour. You have to live with the product to find the real problems.
Shipping is not the finish line. v0.1.0 was the starting gun. Five days of bug fixes followed. That's the actual work — not the glamorous launch, but the quiet week of making things actually reliable.
The AI can build it, but you have to break it. v0.1.0 was vibe-coded. v0.1.1 was debugged by a human using the actual app, filing concrete bugs, and verifying fixes. The AI wrote the fixes. I found the problems. That's the contract.
What came after the foundation
The next iteration did not become a conventional productivity dashboard. The product became more opinionated:
- every local calendar date is its own task workspace;
- Today is interactive, future dates are for planning, and past dates are read-only history;
- target capacity stops the plan from promising more time than the day contains;
- timers stop at their scheduled midnight and respect DST boundaries;
- unfinished work moves only through explicit Carry or Repeat actions;
- completed work stays inline with a green check and strikethrough;
- selected-day or full-history CSV export replaces a heavyweight statistics screen;
- a stable 340 pt, content-sized popover and one shared view model replaced the fragile focus/layout cycle;
- 66 tests now attack rapid saves, hostile names, huge lists, calendar transitions, sleep continuity, carry atomicity, and rendering stability.
Notifications, cloud sync, global timer shortcuts, and trend dashboards never made the cut. Code signing remains future distribution work. v0.1.1 was the foundation; the final build is the product that foundation made possible.
Still 19 Swift files. Still zero dependencies. Still built by someone who never shipped Swift before this month.
Previously: The Problem FocusStation Solves → · v0.1.0 First Release →