How Office Coffee Ratings Work: Deskbrew's 1–10 Score Explained

Most workplace review sites collapse a dozen unrelated judgments into a single star. Salary, management, the commute, and the coffee all land in one number, which makes the number useless for any specific question.

Deskbrew asks one question: how good did the coffee taste? A rating is a whole number from 1 to 10, optionally with a reaction of at most 160 characters. That constraint is the point. A narrow question produces comparable answers, and comparable answers are the only ones worth ranking.

Free coffee is not better coffee. A subsidised bean-to-cup machine can pour something worse than a paid filter brew down the corridor, and an office that stocks four kinds of milk can still over-extract every shot.

So cost, milk options, machine type, and visitor access are recorded as facts and exposed as filters. You can ask for offices with free coffee, or with oat milk, and then sort those by taste. What you cannot do is earn a higher score by buying more equipment.

• Cost — free, subsidised, or paid
• Milk — fresh, oat, lactose-free, soy, almond, or none
• Machine — bean-to-cup, traditional espresso, capsule, filter, or staffed barista
• Access — employees only, by appointment, or open to visitors

A raw mean is trivially gamed. One 10 from one enthusiastic person produces a perfect score, and a perfect score outranks an office with forty ratings averaging 8.4. Every ranking that uses a raw mean eventually ranks its own noise.

Deskbrew applies a Bayesian prior instead. Each office starts pulled toward a neutral 6.5, and real ratings drag it away from that centre in proportion to how many there are. Ten ratings move a score meaningfully. One barely moves it at all.

The practical effect is that climbing the ranking requires genuine, repeated agreement. There is no shortcut where a handful of coordinated ratings vault an office to the top, because the prior discounts exactly the thin samples such a campaign produces.

A score computed from four ratings and a score computed from four hundred are different objects, and showing them in the same typeface is a small lie. Deskbrew labels each office with a confidence state and holds provisional offices out of official rankings entirely.

That means a new office can show a score immediately — contribution should feel like it did something — while the ranking itself stays honest. The two needs are separable, and treating them separately is what keeps the leaderboard meaningful.

An office that replaced its capsule machine with a proper grinder last quarter is not the office that was rated two years ago. Ratings age, and the score weights recent opinion more heavily so the number tracks the current pour rather than an institutional memory.

Factual attributes carry their own freshness state for the same reason. When nobody has confirmed the milk situation in a long time, the page says so instead of presenting a stale claim with false confidence.