A prompt can generate a page that looks like an event-design site. It can't tell you that a client's real pricing is buried under a dead URL, or that a logo needs a second, simplified version to survive a browser tab, or that a color pairing that reads fine on a calibrated monitor actually fails real accessibility math. Those aren't things you automate past. They're the actual work. This is the honest version of what that work looked like for Bella Passion Creations, broken part left in.
The site I actually inherited
Bella Passion Creations is a real event design business — weddings, milestone celebrations, corporate galas — run by a founder who's been doing this since 2020. The original site was a default Wix build, and the audit I ran before touching anything turned up the kind of problems that quietly cost a business real inquiries: no phone number anywhere, just a buried email link. Real pricing — a genuinely good Silver/Gold/Platinum breakdown — sitting under an unrelated URL slug, linked from nowhere on the site. One testimonial carrying the entire trust story. None of that needed a redesign. It needed someone to actually look, page by page, at what was really there.
Design decisions that came from the business, not a template
The structure followed how event planning actually works for this client, not a generic hub-and-spoke stamped onto any business — services became grab-and-go rentals, tiered packages, and event bundles, because that's genuinely how she prices and delivers the work. The visual identity came straight from her own event photography: a parchment-and-brass palette pulled from a real gold tablescape she'd styled, and a dark "candlelight" section built around an actual photo from one of her events. Nothing borrowed from a template library, because a template doesn't know what her actual work looks like.
The favicon turned into a real design problem worth understanding, not just fixing. Her real logo is a fine-stroke script mark — beautiful at header size, completely illegible once scaled down to a browser tab icon. Cropping tighter just made the strokes thinner. The right answer wasn't to force the real logo smaller; it was to sample the exact tone from her actual artwork and build a simplified monogram in that authentic color, used only at the sizes where the full logo can't survive. Brand fidelity and legibility aren't the same constraint, and treating them as interchangeable is exactly the kind of shortcut a generic build takes without anyone noticing.
I'd also built the whole early identity system around an invented monogram, because I didn't have her real logo yet. When she sent it over, the right move was to drop mine everywhere it mattered, not defend the work I'd already done. That wasn't a design note. It was the business owner telling me whose site this actually is — and listening to that is the part no prompt does for you.
The two bugs that actually mattered
The client sent a screenshot: the mobile footer was overflowing horizontally, whole words clipped off the edge. The cause was CSS Grid — a bare 1fr track has an implicit minimum width equal to its content, and a long word that can't wrap forces the whole grid wider than the viewport. The fix was one word: minmax(0, 1fr). I'd built footers like this before and never hit it, because I'd never had a real client send me a real screenshot from a real phone.
The brass-on-parchment palette looked completely readable on my screen. Then I actually ran the WCAG luminance math instead of eyeballing it: 3.55:1 on the label text, 3.9:1 on the primary buttons, both failing the 4.5:1 threshold. None of that was visible by looking. It only showed up once I stopped trusting my own eyes and did the arithmetic.
The bugs that make it into a blog post are never the ones you expected. They're the ones a real screenshot from a real phone catches that your own testing never would.
What actually shipped
By the end, the comparison wasn't close: a real phone number and contact path everywhere the old site had nothing, pricing two clicks from anywhere instead of buried under a random URL, and her real logo and real community partners given the same generous treatment the original site gave them. I graded the rebuild the same way I'd grade a client's live site — B+, 86 out of 100, not a victory lap. No working form backend yet, unoptimized image weight, and legal copy that still needs an attorney's eyes are still open. A rebuild isn't done when it looks finished. It's done when the punch list is empty.
None of the sixteen pages are what make this site good on their own. What makes it good is that the pricing structure matches how she actually sells, the palette comes from her actual work, and the favicon exists because someone noticed her real logo couldn't survive at 16 pixels and cared enough to fix that instead of shipping it broken. That's the part a generic build skips.