How Fynder ranks

01

The Fynder Score

One question, scored. Every place Fynder considers gets a Fynder Score — a decision score from 0 to 100, computed fresh for your mode, your location, and this exact moment. It answers one question:

From where you stand, with what you need, right now — how good a decision is it to go here?

You never see the number. You see the reason: "open when you arrive," "a short walk, highly trusted." A raw score asks you to trust math. A reason lets you check it.

First, the gates

Before anything is scored, a place has to pass hard gates: it must exist and be operating, match the mode you chose, and — if you asked for open places only — actually be open. In Emergency, the facility type must match the help you asked for. A place that fails a gate isn't ranked low. It isn't ranked at all.

What feeds the score

Real travel time. Straight-line distance lies, and in Galata it lies uphill. A few hundred meters on the map can be a staircase. Fynder ranks on routed travel time for your travel mode — walking or driving — not on how close a pin looks.

Patience is modeled per need: the value of a place falls with every added minute, and the rate is tuned to the mode — you'll walk farther for a museum than for a coffee, and in Emergency, minutes dominate everything. When live routing fails, Fynder falls back to an estimate and marks it as approximate — an estimate is never dressed up as a measurement.

Open when you arrive. Not "open now" — open when you get there. This is its own layer, explained in full below.

Category fit. In Eat mode, a hotel with a restaurant is still a hotel. A place whose primary identity doesn't match your intent loses category fit and can't take the top spot from a true venue — it may still appear, lower, honestly labeled as what it is.

And, in small, capped doses:

• Editorial signal — the human İstanbul layer, capped at 12% of the score. Its own section below.
• Your preferences — if you let Fynder learn, your tastes nudge the ranking by a small, capped amount that starts near zero. Never in Emergency. You can see, edit, or reset all of it in Settings under What Fynder learns.
• Data quality — a place with confirmed hours, a verified phone, and consistent data edges out one full of gaps.

Penalties — and where they come from

Scores can only be inflated if nothing pushes back. So penalties push back: closing soon, thin data, a duplicate branch of the same chain, an ambiguous category, imprecise routing, a stale editorial note, a suspected rating anomaly. Every penalty lives in versioned configuration — not vibes — and every ranking change ships under a version number. Same place, same spot, same moment, same data: same answer. The ranking is deterministic. When it changes, it changes on purpose.

The blend changes with the need

There is no single global formula. Eat leans hardest on trusted quality. Explore leans on travel time and how clearly a place is a landmark — a monument next to you shouldn't vanish just because it has fewer reviews than a famous museum across the water. Fun weighs the hour heavily. And Emergency ignores ratings, editorial taste, and your preferences entirely: travel time and verified openness decide, full stop.

02

Trusted quality — why a 5.0 can lose to a 4.7

A 5.0 rating from three reviews is a promise. A 4.7 from thousands is a record.

Fynder uses a Bayesian idea, in plain words: a rating only counts as far as its reviews have earned it. With a handful of reviews, a place's effective rating leans on the average for its area and category — because three delighted friends can produce a 5.0, and time hasn't tested it yet. As the review count grows, the rating gradually stands on its own. Review volume itself also counts, on a curve of diminishing returns: the difference between a few dozen reviews and a couple of thousand matters a lot; beyond that, more reviews add little.

So near Galata Tower, a side-street café holding 5.0 from a handful of reviews starts behind a Karaköy lokanta holding 4.7 across thousands — until it earns its confidence. And a brand-new place isn't punished for being new: it's labeled honestly — New / not enough data — and simply can't take the top role until its record grows.

One thing we never do: claim to read every review. Google supplies the rating, the review count, and at most a few sample reviews, which we show with full author attribution. What we rank on is review confidence — how far the volume lets us trust the rating. Nothing more is claimed, because nothing more is known.

03

Three roles, not a leaderboard

A ranked list answers "what's around?" A decision needs "what should I do?"

That's why the top of every Fynder list is framed as three roles — three different jobs, not positions one to three:

Roles beat lists because a list makes you do the comparing. Position two on a leaderboard tells you nothing — is it almost first, or half as good? A role tells you what trade you're making: the surest bet, the faster door, or the more characterful detour. Three framed answers end the scroll; twenty ranked pins restart it.

Rules the roles obey:

• Each role must earn its slot. The Discovery pick clears the same quality floor as everything else — "different" is never an excuse for "worse."
• No echoes. The three picks can't be branches of the same chain or copies of the same experience. If the best and the closest are the same place, the runner-up steps up.
• Three is a frame, not a quota. Below the roles sits the full ranked list — every place that clears the quality bar, uncapped. Fynder never trims a noteworthy place to keep things tidy, and never pads with a weak one to fill a slot.

04

The Right Now Layer

"Open now" answers the wrong question. You aren't there now. You'll be there after the walk.

So for every candidate, Fynder computes your arrival:

The buffer absorbs real life — the wrong turn, the photo stop — and it varies by need: on the order of ten minutes for a meal, longer for a museum, in between for a night venue. These are starting values, tuned with real usage. Emergency uses no buffer at all: only the freshest travel time, recomputed whenever you return to the app or move meaningfully.

Every place is then checked against that arrival moment and lands in one of five states — shown in the app exactly like this:

• Open when you arrive
• Closing soon
• Opens soon
• Closed
• Hours not confirmed

• Enough time, not just an unlocked door. A museum that's technically open at your arrival but closes too soon for a meaningful visit doesn't get recommended as if nothing were wrong.
• Freshness. The arrival math is short-lived by design. Come back to the app after a while, or move down the hill, and it's recomputed — an answer about right now is not allowed to quietly age.

05

Honest emptiness

Sometimes the truthful answer is: not here, not now.

Late on a quiet night below the tower, if nothing within walking range clears the quality bar, Fynder will not dress up a weak option as a find. The app says, app-verbatim:

The rules behind that moment:

• Two strong results → you see two. The third slot is not filled with a weak place wearing a confident badge.
• One strong result → you see one, clearly framed, with the offer to search wider.
• None → the empty state, with a suggested wider radius. In Emergency, Fynder widens the search automatically and immediately — no one should have to tap "search wider" for help.
• The honest divider. Below the picks, ordinary open places that don't merit a recommendation appear under a plain divider — More open nearby — not picks. Listed, never dressed up.
• The missing are accounted for. When closed places thin the list, the app says so — "… nearby places are closed right now" — so a short list reads as a fact about the hour, not a failure of the search.

An empty state costs us a tap. A wrong recommendation costs you an evening. We know which is more expensive.

06

The City Quality Layer — and its 12% cap

Platform data is wide but it isn't a local. It doesn't know that a lokanta is a neighborhood institution, that a new opening is genuinely promising, or that a viewpoint café coasts on its terrace. The City Quality Layer adds that knowledge — human, editorial, and İstanbul-first.

What an editorial record is. A short original note, tags like "Local institution," "Hidden gem," "Worth the detour," a neighborhood placement (Karaköy/Galata is one of the launch clusters), a verification date, and the editor's audit trail. Every record ages: what isn't re-verified goes stale, and stale records stop counting. Only records verified by the team — or, within limits, confirmed via a vetted local source — affect ranking at all.

It means an editor's affection can never:

• turn a closed place into an open one,
• paper over weak review confidence,
• drag a far-away place into "Best overall,"
• touch Emergency results in any way,
• or disguise a sponsored relationship as taste. Editorial signal is never for sale.

Hidden gems, honestly. "Hidden gem" is not a synonym for "obscure." A place earns the tag only by clearing the quality bar, being verified recently, and offering something a chain can't — and the tag nudges within the cap; it never overrules the record.

Factual references, not partnerships. Where a public selection exists — for example the MICHELIN Guide Türkiye 2026 selection — Fynder may show it as a labeled, sourced fact. A reference is information, not a ranking deal.

The "Viral now" shelf follows the same discipline: an editorial, curated selection with source and verification date. Fynder does not collect or scrape data from TikTok or any social platform — and virality expires, so stale viral tags drop automatically.

07

Trap risk — the score drops, the reason shows

Some places are engineered for people who will never come back. İstanbul has its share. Fynder handles them with a rule we can defend out loud:

A trap-risk place is never hidden. Its score drops, and the reason is shown.

• The "tourist trap risk" tag is editorial and verified — it comes from field-checked curation, never from an automated guess, and like every editorial record it carries a verification date and expires if not re-confirmed.
• The penalty is capped, deliberately smaller than the editorial ceiling — a warning can lower a place, but no single flag can silently sink it off the list.
• The place stays visible, with the badge on the card and the reason spelled out under Why this one. You see the place, and you see the trap. Your call — informed.
• The tag never feeds positive signal anywhere, and none of this applies in Emergency: taste and warnings play no part in routing help.

Silently hiding a place would make Fynder unfalsifiable — you can't question a ranking you can't see. A visible penalty with a visible reason keeps us accountable to you.

08

What Fynder does not do

The shortest section, and the one we most want held against us.