Search the dish, not the restaurant.

Why this exists

You are hungry and you know what you want. You want ramen, or a banh mi, or a birria taco. What you do next is type the name of the dish into a search engine and get back a list of restaurants, sorted by an average over their service, their parking, their room and every other dish on the menu.

That average has thrown away the only thing you were shopping on. A 4.6-star restaurant can serve a forgettable burrito, and the best burrito in the county is very often at a 3.9-star place with slow service and nowhere to park. Star ratings are not wrong so much as aimed at a different question — and once a number has been averaged over a whole restaurant, no amount of arithmetic afterwards can recover which of its dishes is the one worth crossing town for.

So Forkelo ranks the dish. Not the restaurant that serves it.

What Forkelo actually is

Three things, in this order.

A food graph. Canonical dishes — the concept of a carne asada burrito, with all its aliases — connected to real menu items at real restaurants, with prices, addresses and coordinates. One restaurant's “Super Asada Burrito” and another's “Burrito de Carne Asada” are the same concept and belong in the same comparison.

A body of evidence. For each of those dishes, a file of individual, dated, dish-specific mentions gathered from review sites, community discussions, food blogs, editorial writing and Forkelo's own users. Each one is recorded as a structured observation — what it said, how emphatically, how recently, and how sure the researcher was that it referred to this dish at this restaurant.

A score, and a ranking. Those mentions add up to a number between 0 and 10 by a published formula, with a separate, honest label for how much evidence stands behind it. Then the dishes that share a concept in a place are sorted, and that is a ranking page.

The whole calculation is written out, constant by constant, on the methodology page — including the parts that are judgement calls and the places where it is known to be weak. A score nobody can take apart is a score nobody should trust.

That page also sets the Forkelo Score against the film-consensus scores it is most often compared to, and says where it parts company with them on purpose: it weighs how strongly something is recommended rather than counting votes, it lets old evidence fade because a kitchen is never finished, and it prints how sure it is instead of quietly withholding a badge.

How a ranking gets built

Depth beats breadth: twenty-five well-researched burritos are worth more than three hundred thin ones. One dish, in one city, at a time.

1. Pick the dish and the placeOften from what people searched for and got nothing back — an unanswered search is the clearest signal that a ranking is missing.
2. Find every contenderRestaurants in that city plausibly serving it, usually twenty-five to thirty to start. Not all of them survive.
3. Map the menu item to the conceptThe real menu name, with its real price, linked to the canonical dish so it lands in the right comparison.
4. Research the dish's reputationThe actual work. Find what people have said about this dish specifically — praise, criticism and shrugs alike — and file each source as its own row, in the researcher's words, never copied.
5. Let the score fall outNobody sets a score by hand. Each new piece of evidence recomputes it from the whole file.
6. Send anything uncertain to a personA restaurant that looks closed, an ambiguous menu mapping, a possible duplicate, a conclusion drawn from thin evidence: all of it queues for human review instead of going live.

A ranking is published only once at least three dishes qualify. Below that it is a list, not a comparison, and a page that pretends otherwise is worse than no page.

What we hold to

• The dish, alwaysIf a claim is about the restaurant rather than the food, it is not evidence and it never enters a score.
• Show the workingEvery score decomposes into the mentions that produced it, and the method is published in full rather than described in adjectives.
• Say when we don't knowQuality and certainty are different questions. A thinly evidenced dish is marked “Early ranking” and stays that way until somebody does the work.
• Old news gets quieterKitchens change. Evidence decays with a half-life, so a dish cannot coast forever on a reputation it earned in 2021.
• Criticism is not a failureA database of only compliments produces rankings nobody believes. Negative and neutral mentions are filed as carefully as praise.
• No button that liesNothing on the site offers to do something it cannot do, and no page exists that has nothing on it.

How this is paid for

Not yet, is the honest answer — Forkelo is early access and is being built before it earns anything. But the boundary is fixed now rather than later, because it is the kind of thing that is easy to hold to before there is money on the table and impossible afterwards:

• A rank is not for sale. Not a position, not a placement in a ranking, not a tie-break, not a badge that looks like one. There is no field in the database a payment could set.
• No advertising on ranking pages, and no sponsored rows dressed as results.
• No third-party tracking. Forkelo counts a handful of clicks in its own database to check whether the premise works, honours Do Not Track and Global Privacy Control, and loads no third-party scripts. The privacy page maps every claim to the code that makes it true.

What a restaurant can do is correct a fact — that it has moved, changed its hours, or taken a dish off the menu. What it cannot do is move its dish.

When we get it wrong

Sometimes a ranking will be wrong, and usually it will be wrong in the same way: a genuinely great dish that nobody has written about is invisible to a system built on what people have written down. That is the known weak point, and the fix is a person telling us where to look.

Every ranking page carries a “Tell us what's better” form. What you send goes to a researcher, not to the score. Nobody can move a number by filling in a form — what happens is that somebody goes and researches the restaurant you named, files whatever they actually find, and the score moves on its own or does not. Several people naming the same place is a strong signal that a ranking has a hole in it.

For anything factual — a restaurant that has closed, moved, or renamed, or a dish that is off the menu — email support@forkelo.com and it will be corrected.

Who makes it

Forkelo is built by Vorby Studios LLC, a small studio in the United States. It is launching in San Jose and the wider South Bay — one market, done properly, before a second one — because a directory that is thin everywhere is useful nowhere.

If you want to be told when it opens, there is a list. If you run a restaurant, there is a page for that. If you want to know exactly how the number is computed, that is the long version.