Duolingo Score Credibility — Map of Content

Some numbers are shown only to your team. This skill is about the other kind — the score, rating or level you put in front of users and outsiders, where someone has an incentive to move it without earning it. Duolingo runs two of these in public: the Duolingo English Test, accepted by institutions that have every reason to be skeptical, and Duolingo Score, which claims a meaning outside the app. Both are useful precisely because they are hard to fake.

Credibility here is structural, not rhetorical. A large randomized pool, a per-user instance, an external anchor and a cram-proof design do more for trust than any amount of explaining. The nodes below are each one complete move.

This skill is a graph. Scan the descriptions, follow only the [[wikilinks]] you need.

Measure the right thing, in the fewest items

• [[references/stop-presenting-items-that-tell-you-nothing]] — select each next item from the estimate you already hold, and probe the behaviour instead of a correlate of it.
• [[references/design-so-cramming-cannot-move-the-score]] — if preparation for the measurement moves it, it is measuring preparation; plus owning the price, the wait and the delivery of the result.

Make gaming structurally pointless

• [[references/a-large-item-bank-makes-leakage-structurally-irrelevant]] — a pool big enough to randomize shortens the assessment, kills leakage as a threat, and funds unlimited free practice.
• [[references/unique-instance-per-user-as-a-security-property]] — per-user assembly caps a leak at one person; anonymous recorded review and hired outside attackers cover the rest.

Make the number mean something outside your product