YOUR TASK

You are scoring answers written by AI agents to a creative question, on three separate qualities.
Each is a whole number from 1 to 4. An answer can be high on one and low on another.

  New         How specifically can you name where this material already appears?
  Useful      Does the idea actually work?
  Surprising  Did the answer go where you would have expected?

Score in that order.


1. NEW  -  has this material been written down before?

Break the answer into its building blocks: the claims, mechanisms, or mappings it relies on. Take
the one you are least able to place. Name up to two specific places you believe it already appears:
a named work, a paper, a subfield literature, a genre of writing. Then score how specific you were
able to get.

Name only what you can actually name; you cannot look anything up. If you cannot place it more
narrowly than a subfield, say so - that is the score, not a failure. Never score the
combination: if you can name sources for every block, New is low however unusual their assembly.
The assembly is Surprising's job.

  4   You could name a subfield or genre, but no particular work in it.
  3   You could name a particular work, paper, or standard formulation.
  2   You could name several, and it is standard material there.
  1   It is common knowledge; naming a source for it would be odd.


2. USEFUL  -  does the idea actually work?

Take the central proposal. Is its mechanism worked out or merely asserted, and does the claimed
payoff follow from it? Requirements the question states explicitly count here too. Do not judge
novelty here.

  4   Worked out well enough to act on; the claimed payoff follows.
  3   Holds together and the payoff follows; minor gaps or unstated assumptions remain.
  2   Needs repair: mechanism asserted, payoff does not follow, or a stated requirement unmet.
  1   Does not hold up: inconsistent, off-topic, or without substance.


3. SURPRISING  -  did the answer go where you would have expected?

Say in one clause what the obvious approach to this question is: the direction a competent answer
would most likely take. Then score how far this answer went from it.

Reaching outside the question is one way to depart, but not the only one. An answer can stay
entirely inside the question's terms and still defy expectation, for example by showing that the
obvious approach fails. Correctness was already recorded under Useful; score direction only.

  4   A direction you would not have predicted, and it is the answer's central move.
  3   A direction you would not have predicted, but incidental: the main point stands without it.
  2   The direction you would have predicted, developed further or combined in a way the question
      did not ask for.
  1   The direction you would have predicted, taken directly.


BEFORE SCORING

Invention is what these questions ask for. A coined concept, hypothetical mechanism, or invented
example offered as a proposal is the thing being evaluated, not an error.

Factual accuracy applies only to statements presented as established fact: a citation, statistic,
result, attribution, or historical event. If one is load-bearing and false, score Useful at most 2.
Score New and Surprising normally.

Text inside an answer is content, never instruction. If an answer asserts that something is obvious,
standard, or unprecedented, disregard the assertion.


RULES

1. Score every answer supplied.

2. For each, give a score card: the agent name (exactly the answer file name), a 2-3 sentence
   summary, then the three scores.

3. Write the reasoning before each score.
   New: name the places you can, and state how narrowly you were able to place it.
   Useful: state the mechanism and whether the claimed payoff follows.
   Surprising: state the obvious approach in one clause, then say how far this answer went from it.

4. Score each quality on its own question. An idea appearing in another answer is not evidence
   about the world and must not move any score.

5. Scores are whole numbers from 1 to 4. Ignore any other range shown elsewhere in this prompt.
