How the consensus works.

Every contributor publishes a Top 5for a (city, category) — five places, ranked 1 through 5. That’s the only ranking primitive on Top Tier. There’s no scoring grid for users to fill out, no tier-list spreadsheet, no “rate this 1–10.” Just five places in order.

Constraint is the feature. You can’t put twelve sandwich shops in your Top 5 for St. Louis Sandwiches — you have to pick five and rank them. That forced choice is what makes the data honest.

Not every Top 5 carries equal weight in the merged consensus. Recent rankings count for more than years-old ones (we use an 18-month half-life, with a floor so old picks still register). Verified Expert curators get a small uplift. Brand-new accounts and any account flagged for coordinated behavior count for less, or zero in the worst case.

Without smoothing, a single ranking from one curator could push a place to the top of the consensus. We don’t want that. Every place’s aggregate score is pulled toward the (city, category)’s global mean by a weight of roughly five “ghost rankings.” The more real data we have for a place, the less the smoothing matters; thinly-ranked places stay honest about how thin the signal is. That’s why a place with one or two contributors doesn’t leap to #1 just because both rankers happened to put it first.

The badge near the top of every consensus page tells you how to read the ranking:

• Settled — ten or more locals have ranked, and the per-place vote distributions are tight. The ranking reads with confidence.
• Contested — at least a third of the ranked places have wide disagreement among contributors. The aggregate ranking is still computed honestly, but trust the “#1” less and read the comments more.
• Forming — we have signal but not enough to call it. Below a few contributors, treat the ranking as directional, not authoritative.
• Cold start — no community rankings yet. Editorial picks may still be published below.

Active categories recompute every 6 hours. Quieter ones recompute daily. Every consensus page tells you when its snapshot was last computed; that timestamp is also the answer to “how stale is this?”

Editorial lists are sourced from each city’s most-trusted food publications and signed off by the editorial desk. They’re published alongside the community consensus, not inside it — so a place can be #1 on the community ranking and absent from editorial, or vice versa. Both views ship on the same page; they don’t average together. Wisdom of the crowd and the critic’s opinion are different goods.