Why Community-Driven Discovery Works

A decade ago there was a prediction about discovery: the future would be algorithmic. AI would figure out what you liked and surface it. Looking back, half of that prediction came true — algorithms are everywhere now. The other half didn't: discovery quality didn't really improve. Spotify's weekly recommendations felt magical for a while; now they're predictions from your 30-second listen patterns. YouTube recommendations are optimised for keeping you on the screen, not for teaching you. Are we trusting TikTok to decide what video to watch next?

Inside that landscape, against expectation, human curation didn't die; the old venues are still standing. Hacker News, Reddit, Product Hunt, Indie Hackers — all built on human voting at the core. In a battle algorithms were supposed to win, small communities held quality.

Building Omnifetch as community-driven was a deliberate choice, not an accident.

What algorithms actually optimise for

A point that's often skipped: algorithmic recommendation systems aren't optimising for quality; they're optimising for engagement. "Did the user find this content good" is hard to measure; "how many seconds did they watch, how many times did they scroll, did they comment" is easy. The system optimises what it can measure.

The two often overlap. But not always. Human brains engage more with anger, fear, sensation. So an engagement-optimised system pushes manipulative, polarising, sensational content over time — even when the user doesn't want it.

Same dynamic in product discovery: products with flashy landing pages, exaggerated promises, AI-buzzword descriptions get clicked. The algorithm flags them as "popular," shows them more, and crowds out the genuinely useful but modestly-presented ones.

The different incentive structure of community curation

Community curation runs on a different mechanism: reputation.

When you upvote a story on Hacker News, you're not just casting a vote; your account is behind it. Accounts that consistently upvote irrelevant or low-quality items get less effective weight over time, and other users notice. There's an unconscious "how does upvoting this make me look" question behind every vote.

That social incentive is something algorithms can't replicate: the human desire to be seen as someone with good taste. Algorithms optimise engagement; humans optimise being-respected. Different things.

The practical advantages

First, the spam and manipulation surface. Manipulating algorithms with bots is a well-developed industry. Manipulating an active community is harder, because account age, posting patterns, and comment quality are all signals that humans evaluate.

Second, feedback richness. An algorithm sees one bit: clicked or not. A community member comments, critiques, suggests alternatives, asks. The value of that feedback to whoever shipped the product isn't comparable to a click-through rate.

Third, context. "This library is great" plus "but the TypeScript support is half-baked, for X I'd pick Y" is multiplicatively more useful than the first sentence alone. Community curation carries context; algorithmic recommendation doesn't.

Fourth, niche depth. Algorithms amplify what's already popular. Communities can go deep into niche interests. A small subgroup focused on "Rust-written CLI tools" can produce a level of curation no algorithmic feed will reach.

What it takes for it to work

Saying community curation just works is wrong. Badly-designed communities (see: any subreddit that grew too fast) start behaving like algorithmic feeds. For the dynamic to hold, a few conditions matter:

Bounded size, or good subdivision. Reddit scales by splitting into subs; HN stays one feed at roughly 10x smaller and quality holds. Slow growth — communities that 10x in a week lose quality. Editorial backstop: the community doesn't decide everything; intervention against problem posts and rule enforcement matter. Pure "democratic" communities tend to corrode. Reputation mechanism: if a five-year active account has the same vote weight as a brand-new one, brigading is trivial.

Where Omnifetch is right now

I want to be honest about what we've built, and what we haven't:

Right now we have a simple voting system: one user, one vote per product. Account age and history don't yet weight that vote — we haven't built it. Spam will become a real problem at a certain user-count threshold; the plan is to add reputation-based weighting then. But framing it as "already implemented" would be misleading.

There's an editorial "Featured" layer. Trending is community-chosen; Featured is editorial — quality plus ecosystem representation. The two layers complement each other, they don't conflict.

Comments live inside the product page. The comment section makes the context behind a product visible. "Who used this and for what" lives not just in the product description but in the discussion the community builds around it.

Collections are a form of mini-curation. Anyone can build and share a collection. That creates parallel curators across the ecosystem; active ones will get editorial promotion later.

Closing

I'm not saying algorithmic discovery is dead. I'm saying it doesn't work where something better is possible. Human communities are one of the rare areas where quality doesn't have to give way to scale. As Turkey's maker ecosystem grows, building discovery infrastructure on this logic — algorithms as support, human reputation as the core — is the more sustainable path long-term.

If that logic appeals to you, join in. A vote, a comment, a collection — each is a small piece of the ecosystem's discovery infrastructure.