Is this topic good enough to reach the finals?
Part of a nine-skill suite for the S.T. Yau High School Science Award. It either proposes a shortlist of topics from a student's profile, or scores a topic the student already has — grounded in a local archive of past winners and the award whitepaper, and weighted toward the one thing Yau judges reward above all: original work the student can defend as their own.
The S.T. Yau award's governing philosophy is 自己出题自己做 — you pose
your own problem and you solve it yourself. yau-topic-finder
runs in two modes. In discover, it turns a student profile into a
ranked shortlist of candidate topics across the award's six subjects, each with
a confidence-banded score. In score, it takes a topic the student
already proposes and returns a scored breakdown, an originality-and-ownership
diagnosis, and concrete pivots to make it stronger. Every judgement is grounded
in a local archive of 393 winner entries (2020–2025) and the whitepaper — and
it never invents a winner, a paper, or a citation.
In
discover: a student profile — subject, background, interests, constraintsscore "topic text": a proposed research topic- Optional subject flag (math / physics / chemistry / biology / CS / econ)
Out
- A ranked topic shortlist, or a scored breakdown of one topic
- Confidence bands — scores as ranges, never false precision
- An originality & student-ownership diagnosis
- Pivot suggestions to raise the ceiling
Scoring a proposed topic
A student proposes a topic; the skill scores it and says, plainly, where it is strong and where it will struggle in front of the panel. The exchange below is illustrative.
/yau-topic-finder score "Using a convolutional neural network to
classify handwritten digits on the MNIST dataset" --subject csDiagnosis. MNIST classification is a solved, textbook benchmark — feasible, but a CS panel has seen it hundreds of times. There is nothing here the student can claim as their own question.
Pivot. Keep the student's real interest (image models) but move to a problem they can own: a small, self-collected dataset from their own environment, a failure mode they noticed and can characterize, or a method comparison that answers a question they posed. Originality and ownership are what move a Yau topic from "fine" to "finals".
Bars and labels are illustrative; the real skill reports confidence ranges, not single numbers.