Early on, every new person who joined us lost the same forty minutes to the same ghost. They would follow our own written instructions to the letter, type in exactly the address we told them to use, and get a wall of nothing. They would check it twice, blame themselves, and sit there stewing until someone leaned over their shoulder and said the magic words: oh, ignore the instructions, the real door is over here. The instructions were wrong. They had been wrong for as long as anyone could remember. And the sign-in itself kept dying at the threshold, because a piece the whole thing depended on was never actually being switched on at start-up. Two unrelated-looking little fires, both burning quietly in plain sight.
The obvious instinct, when something misbehaves, is to leave the structure alone and just swat the bug. Fix the wrong address. Add the missing piece to the start-up list. Move on. Instead we did the opposite of what every architecture talk on the internet preaches. We took two pieces we had filed under optional and peripheral and dragged them right into the beating heart of the product. More weight in the critical path. More things that absolutely must be alive for the product to feel alive. A bigger crater if any of it falls over. The whole internet tells you to push things to the edges and keep the center tiny. So why pull them in?
Because the optional label was already a lie. Search was never optional. The thing that arranges everything on your screen was never optional. If either went dark, a customer saw a broken page and reached for the phone. Calling them optional just meant they got optional care: thinner safety nets, fuzzy ownership, no reserved seat in the start-up routine. The missing sign-in piece had the same disease. It was a service, not the thing without which nothing works, so it was easy to forget to switch on. The wrong address in the manual was not really a typo. It is what happens when nobody is sure which door matters because nobody is sure which piece matters.
So we redrew the line. We hauled those two pieces into the core, where they now live and breathe on the same clock, with the same safety nets, and the same blunt status as the sign-in itself: if this is broken, the product is broken. The forgotten piece got a permanent seat at start-up, so it wakes with everything else. The address got pinned down to one true answer in one place, and everything around it was rewritten to assume that one truth. One heart. One start-up. One door. One honest answer.
Here is the lesson, and it lands whether you run a software team or a body shop. The tools you call core should be exactly the tools whose absence stops the business cold. Not the ones your vendor calls premium. Not the ones that were expensive. Not the ones you happened to set up first. The ones that, if someone quietly switched them off at nine in the morning, would have your phone ringing by ten. Most operators have this list wrong by at least one item, usually two. A humble scheduling tool gets treated as optional because it lives in a browser tab, while a fancy expensive system gets treated as core because of what it costs. Then the scheduling tab dies and the whole day stops, and nobody can explain why the optional thing took the business down.
Walk your own stack and ask one question per tool. If this vanished at the start of tomorrow, would the front desk notice within the hour? If yes, it is core, and it deserves core treatment: a named owner, a tested fallback, a known way in, a line on the morning checklist. If no, it really is peripheral and can live in the long tail. The labels you inherited from your vendors, or from the order you happened to buy things in, are almost never the right ones. We learned to stop trusting the labels and start trusting the symptom. The addresses lie. The labels lie. The invoices lie. Only the phone ringing tells the truth.