What Shipped, What Broke, and What Users Actually Paid For

Entry one was the why. This one is the receipts. Here's what actually shipped on Metavulus recently, what broke in front of real people, and the part that genuinely surprised me — what users were willing to pay for versus what I assumed they'd pay for.

What shipped

Three things made it out the door, and they tell a story about where the value actually is.

The bilingual daily digest went live. A tight, every-market-day briefing: what moved in crypto and forex, the key levels, the catalysts to watch. The hard part wasn't the writing — it was making it feel native in both English and Bahasa Indonesia without the Indonesian reading like a translation. Trading terms stay English (liquidity, drawdown, tickers), the connective tissue is natural Bahasa. That balance took more iterations than the actual feature code.

The community forum opened up. Threads, replies, upvotes, a bit of gamification. I went back and forth on whether a solo founder should even attempt a forum — they're famously hard to seed — but a quiet, high-signal trader forum was worth the risk. Early signs are okay: small, but the conversations are real, not spam.

The first Pro features landed behind a paywall. This is the one I had the most anxiety about. Putting a price on something you built is a different kind of exposed than just shipping it for free.

What broke

Plenty. Here are the two that actually reached users, because those are the ones that teach you something.

The locale cookie didn't stick on first visit. New visitors from an Indonesian browser sometimes got served the English shell on their very first page load, before the cookie was set, then flipped to Bahasa on navigation. Jarring, and exactly the kind of bug that erodes trust on a bilingual product where the whole promise is "this was built for you." The fix was making the server-side locale resolver fall back to the accept-language header properly instead of leaning on the cookie too early. Boring fix, real impact.

A Pro page rendered for a split second to logged-out users. A flash of paywalled content before the redirect kicked in. Not a data leak — the API was gated server-side — but a bad look and a reminder that "it's secure" and "it looks secure" are two different bars you have to clear. I moved the gate earlier in the render path so there's no flash.

Neither of these is glamorous. That's the point. Most of what breaks in a real SaaS is small, unsexy, and only visible because actual humans are using it in ways you didn't test.

What users actually paid for

Here's the part I got wrong, and I think it's the most useful thing in this entry.

I assumed people would pay for more — more signals, more coverage, more pairs, more frequency. The classic "stuff the Pro tier with features" instinct. So that's what I leaned the early Pro tier toward: volume.

What people actually responded to was clarity and saved time. The Pro features that got used and renewed weren't the ones that added more noise — they were the ones that removed it. The structured research view that says "here's the level, here's why, here's the invalidation" in one screen got far more engagement than the firehose of additional ideas. People aren't short on trade ideas. They're short on trustworthy, time-saving filtering.

That reframed my whole roadmap. The job isn't "give traders more." It's "give traders less, but the right less, faster." Every Pro feature now has to answer one question: does this save the trader time or sharpen a decision? If it's just more volume, it doesn't ship to Pro.

I won't post revenue figures — partly legal, partly because for a product this young the number isn't the signal, the behavior is. The signal is: the people who paid did it for clarity, not quantity, and a meaningful share came back. That's enough to know which direction to dig.

What the AI agents did here

The agents built most of the digest scaffolding and the forum CRUD fast, which freed me to spend my time on the editorial voice and the pricing decision — the parts that needed a human who understands this specific audience. But the locale bug? That was partly an agent-written assumption (lean on the cookie) that looked fine in code review and only fell apart with real traffic. Good reminder that "passes review" and "survives production" are different tests. More on the agent wins and misses in the next entry.

The honest risk note

As always: this is a research tool, not financial advice, and trading carries real risk. The Pro tier helps you think faster; it does not promise outcomes. Anyone selling you certainty in markets is selling you something else.

Next time: a candid scorecard on the AI agents themselves — what they nailed, what they quietly got wrong, and the bugs I only caught because I read every line.