Research Topic Taste Audit

What problem it solves

Research decisions often fail not from lack of effort, but from premature direction commitment: before several months of work, a researcher has not checked whether a question is important, learnable, defensible, and genuinely suited to the person doing it. This skill turns research taste into an executable audit loop for thesis proposals, direction commitments, and self-review before submission, rather than for tuning hyperparameters or deciding product features.

How it works

It exposes three triggers: 科研品味 for full audit, 科研品味 quick for fast calibration, and 科研品味 calibrate for periodic taste review. The core combines a shared base scan with 7 gates: R1 Lead vs Follow, R2 Novelty/learnability, R3 defensibility, R4 elegance/simplicity, R5 vision-driven origin, R6 feasible attack path, and R7 person-topic fit. Each gate is marked red/yellow/green and reweighted by project stage. Full mode outputs the base scan, gate dashboard, per-gate review, counterquestions, final probing, and meta-audit; quick mode compresses the result into a dashboard, verdict, and key counterquestions, which is useful during meetings, chats, or paper reading.

Boundaries and caveats

It is meant for research topic and direction judgment, not for experiment design, coding, baseline selection, or product decisions. The source material is explicit that product-taste auditing has been removed, that insufficient input should trigger clarification, and that a red gate requires stopping and re-auditing. Use it before a direction is fully locked in; do not force every execution detail through the seven-gate model.