Advisory score
An advisory score is an AI-generated evaluation number that informs a human decision but never makes it. It sits next to the human score as a reference point: it can order a pile for reading, flag outliers and explain its own reasoning, but it has no path into the leaderboard, the shortlist or the award.
The distinction is architectural, not a disclaimer. A score is advisory when the system gives it no way to become the result.
A disclaimer is a promise. Architecture is a guarantee.
Most AI screening tools say “human in the loop” and mean that a person may override the machine’s number. That leaves the machine’s number as the default, and defaults win: by submission two hundred, tired reviewers confirm rather than judge. An advisory score inverts the default. The AI number is never the result by construction, the human enters their own score, and the ranking is built from the human number alone.
The same distinction is turning regulatory. Rules such as the ERC AI guidelines prohibit delegating the assessment of merit to an AI system. Advisory by architecture, rather than advisory by promise, is what that boundary looks like in a working product.
Two numbers, one decision.
Every submission in EvalLens carries two numbers. The AI Total Score is the deterministic aggregate of six independent judges: it exists to order the reading, surface disagreement and carry the evidence. The Jury Score is entered by your human judges, and it alone builds the leaderboard. There is no mode in which the AI Total Score ranks anyone, which is the product-level meaning of our line: AI prepares the analysis, people decide.