| name | defense |
| description | Use when simulating Guochuang roadshows, expert questions, rebuttals, speaker scripts, or a final readiness drill for school, provincial, or national rounds. |
Defense
Apply the shared policies in gps-common. When saving a Markdown report, use the common project-name rule with topic 答辩.
Rehearse the questions that can change a reviewer's decision. Challenge the strongest claim and the weakest proof. Answers must come from records, not improvised facts.
Question matrix
Generate questions across need, method, novelty, measurement, customer, economics, IP, team contribution, risk and social value.
Add track questions for red-tour field impact, industry command fit and demonstrator, vocational practice, or sprout autonomy. Add stage pressure for school review, provincial review and national final.
Answer card
Use this order:
direct answer -> evidence pointer -> boundary or limitation -> next action.
Each card also records severity, speaker, screen_evidence, verification_owner and verification_date. If the team does not know, say what is unknown and give a check owner and date. Never hide a material risk or coach a fabricated answer.
Scoring the drill
Score each answer from 0 to 2 on:
- Clarity: the first sentence answers the question.
- Evidence: the pointer can be opened or shown.
- Ownership: the student explains their own work.
- Boundary: limitations and conditions are stated.
A score of 6 or less is a revision item. A missing source or integrity issue is a hard stop regardless of fluency.
Drill modes
Return a severity-ranked question set and choose a mode:
- Script run: timed opening, transitions and close.
- Interrupt run: stop after each claim and ask for its source.
- Adversarial run: combine a strong claim, a missing control and a time limit.
Include the local time limit only when supplied by an official or local notice. The output should contain the script, interrupt points, screen evidence list, failed cards and the next rehearsal date. Passing a drill means understandable and auditable. It does not predict an award.
