Making of — Fleet Mission Control, in plain English

This page grows as the project grows. Companion to ../technical/making-of.md, which has the file:line detail and exact commit hashes; this page tells the same story without needing to read code.

The screen-layout bug (July 2026)

The cockpit's screen rearranges itself depending on how big the window is — three columns on a wide monitor, down to one column on a phone. There's also a special case for foldable phones that are physically bent open into a book or tent shape: when that happens, the screen splits across the two physical halves.

Here's the bug: the code that detects "bent open" was built assuming it would also catch a foldable phone that's opened out completely flat, like a small tablet. It doesn't — a phone lying flat reports itself as one single continuous screen, the same as a normal phone, not as "two segments." So a tablet-sized flat-open foldable was falling through to the cramped one-column phone layout instead of the roomier two-column layout it should have gotten.

What makes this worth telling: everything that could be automatically tested passed. The screenshots looked right at every width someone could actually test on a regular desktop screen. The bug only showed up in a scenario no desktop testing tool can fake — an actual device, actually bent in a specific way. It took someone reviewing the logic itself, checking it against how the underlying phone-hardware feature is documented to work, rather than trusting "the tests are green," to catch it before it shipped to a real folding phone.

The fix didn't touch the bent-open detection at all — that part was already correct for genuinely bent phones. It fixed the plain "how wide is the screen" rule so a flat-open foldable gets caught there instead, which is what should have handled it from the beginning.

The lesson, in one line: a green checkmark means the code didn't crash on what you could test — not that a feature actually works on the specific hardware situation it was built for.

The pick-by-name day (July 18, 2026)

Four related features landed in one day: choosing repos by name instead of typing a folder path, a small family-tree view of related jobs, generating documents and slide decks (not just code changes), and a way to tell a computer "don't send me anything heavy right now."

Two things worth knowing about how that day went:

• The document-generation feature caught a real mismatch with a bit of detective work. Some earlier work had left a test expecting the exact words "view artifact" on a button. A later change that same day made the button say "view my-slides.pptx" instead — more useful, since you can now see the actual filename before clicking. Rather than silently leaving that test broken, whoever built the next feature noticed it, double-checked on a clean, untouched copy of the code that the same test failed there too (so it wasn't something they'd broken), wrote that finding down plainly, and left it for a quick one-line fix right after — which is exactly what happened next.
• A test caught a bug before anyone saw it in the real app. Adding a new "is this computer available right now" folder accidentally made the system think that folder itself was a phantom extra computer, because the code that lists "all computers" was just listing "every folder in this spot." A test written specifically to check the new feature caught the mix-up immediately, and the fix was one line: teach the listing code to skip that one special folder name.

Neither of these is a dramatic story — that's rather the point. A repo with good habits around noticing and writing down small mismatches, instead of looking away from them, tends to produce a string of one-line fixes instead of one big one. The technical version has the exact tickets and commits if you want to trace either story further.