The judge in a box.
A student hands over their poster; the skill reads it, runs parallel research agents to check the science against current literature, scores it against the official ISEF or MSEF rubric, and hands back the questions the judges are likely to ask — with model strong and weak answers. It is the dry run before the real panel.
science-fair-judge takes a finished poster or slide deck and does three things at once. It fans out parallel agents to verify the project's scientific claims against current literature; it grades the work against the official Grand Award judging criteria for the target fair; and it assembles a categorized bank of interview questions — general, background, methodology, statistics, results, limitations, impact, hypothetical — each with the answer a strong student would give and the answer that would lose points. The output is a single Markdown report the student and coach can work through together.
In
- A poster or presentation — PPTX, PDF, or images
- The target fair (ISEF, MSEF, or another rubric)
- Optionally the category / sub-category for panel context
Out
- A literature-accuracy check on the science
- A rubric score with per-dimension feedback
- 60+ categorized judge questions with strong / weak model answers
- One Markdown report with scores, feedback, and prep guidance
A poster goes in; a scored review comes out
Below is an illustrative fragment of the report the skill produces for a hypothetical project — the rubric read, then a sample judge question with the two model answers. The example is invented for illustration; no real student project is shown.
The weakest line — data analysis — becomes the first thing the coaching session addresses. Bars are illustrative.
Q. Your effect looks real, but with a sample of thirty, how do you know it isn't noise? What test did you run, and why that one?
Reveal strong vs weak answer
Strong. Names the test and why it fits the design (paired vs unpaired, normality assumption), reports an effect size and confidence interval rather than only a p-value, and states the sample-size limitation honestly with a plan to address it.
Weak. "It was significant, p < 0.05." No named test, no effect size, no acknowledgement that thirty samples is small — the answer a judge marks down.
The judge review naturally leads into interview drilling, so this skill hands off to a companion, science-fair-interview-prep, which expands the question bank into a full bilingual Q&A document — and to science-fair-prep-feedback, which reads the student's annotated version back and responds question by question. It is one link in Tian2's competition-coaching chain, not a standalone gadget.