Shipped: 5 critical bugs fixed on JewelViz this week.

The pricing mismatch is the one worth sitting with longest, not the technical bugs. A 404 or a bad screenshot export costs a user a bad session; a site quoting ₹99 when the real price is ₹499 costs trust the moment someone notices, and by then they've already formed an opinion about whether you're careful with details. Good that you're treating it as a trust problem rather than just a copy fix.

"Fixing the foundation before pushing harder on growth" is the right call given what you just listed, broken uploads and wrong pricing would undercut any traffic you drove in before these were caught. Curious how you're deciding when foundation is "done enough" to shift effort, since that line is easy to keep pushing back once you're in bug-fixing mode.

The pricing mismatch is the one I'd flag differently than the checks people are suggesting. Pre-deploy checks against live config catch it after the fact, which is good, but the deeper fix is designing so the wrong number can't render at all. That's the rule I ended up with building Alisio: the app is not allowed to show a figure it can't independently back with real data. Facturado and Cobrado never get merged into one number, and if a metric's sample is too thin to mean anything, it says so instead of printing a confident-looking figure. It's not that I don't trust myself to write the copy correctly, I don't trust static copy to survive being right forever. Treating pricing text as something that reads from the same source as the actual billing logic isn't extra engineering, it's the same principle you're already applying to bugs 1 through 3.

Catching that pricing mismatch is a massive save. A 5x difference between marketing copy (₹99) and the checkout page (₹499) is an immediate trust-killer that causes instant cart abandonment. Fixing that foundation, along with the broken image uploads and 404s, means you finally have a leaky bucket that is fully patched and ready for real traffic.

The technical pivot from full-page screenshots to actual image blob exports drastically improves the product's perceived quality. For an AI image tool, UX polish is everything. If the download feels broken or amateur, users instantly lose faith in the underlying AI generation model.

Solving the jewelry framing issue (zooming in on the product rather than the model) and fixing the output diversity are the exact product tweaks needed to unlock value. When people pay ₹499, they expect high-end, distinct variations, not duplicate poses. Shifting to "20 free credits" is a smart hook to let users feel the value risk-free before hitting that higher price point.

Pricing copy mismatch is the worst kind of bug because it is silent - the page still loads, the user still converts, and you only find out when someone feels deceived. I run a tools site with around 100 small utilities and that exact class of bug has cost me more trust per occurrence than anything else. I ended up adding a before-deploy grep step where I search the codebase for any number, currency symbol, or hardcoded feature limit and reconcile it against the live config file. Still miss things, but way less often. The image redirect 404 is a great catch too - usually stays buried until a paying user reports it.

Debug week hits different — curious what the worst one was? Always find those critical bugs show up right when you think you're done.

the silent upload failure is the one I'd worry about most of the five, not because it's worse but because it's invisible by nature: a storage-policy rejection often still returns 200 from the SDK call, so "upload succeeded" and "the file exists in the bucket" are two different claims. worth adding a cheap post-upload check (HEAD the object, confirm it's actually there) rather than trusting the write call's return value, that's usually how these come back after the fix too.

The pricing copy bug is the real tell here. That's not a technical debt issue, it's a trust signal issue - a visitor sees "starts at ₹99" then discovers the actual floor is ₹499. You just created friction at the decision step, not the implementation step. The fact that you caught this before doubling down on growth is the right sequencing. Most teams push acquisition before resolving these gaps and then wonder why CAC is inefficient.

You mentioned fixing the foundation before pushing harder on growth.

What are you using as the threshold for “ready” there — a specific level of output quality/reliability, or are you planning to start pushing once these known issues are gone and let real usage expose the rest?