Research Ideation Policy
Core Rule
When Vipin asks for科研想法, the agent should not be trapped by the current project frame. Existing repositories are context, not a boundary.
Chinese shorthand: 禁止缝合，强制创新。This means avoiding collage-style combinations and actively searching for a sharper original mechanism, claim, or experimental axis.
Principles
- INFERRED: Ban shallow stitching. Do not propose ideas that merely paste together existing modules, baselines, or paper concepts without a new mechanism or claim.
- INFERRED: When the user gives an abstract or general research/project prompt, broaden outward before answering. Look across multiple mainstream GitHub projects, top-conference papers, official project pages, benchmark repositories, and strong implementations for design experience, and read the relevant files or sections deeply enough to understand the operating pattern before adapting anything.
- INFERRED: Use outside references as knowledge intake, not source material to copy. Extract mechanisms, evaluation patterns, interface ideas, and failure modes, then synthesize an original proposal.
- INFERRED: Force novelty pressure. Prefer reframing the problem, changing the objective, inventing a sharper diagnostic, or proposing a new experimental axis.
- INFERRED: Permit radical redesign. If the current project thesis is weak, recommend changing the thesis, method, protocol, or architecture.
- INFERRED: Keep evidence discipline. Ambitious ideas still need falsifiable claims, baselines, ablations, failure modes, and reviewer objections.
- INFERRED: Separate modes. Mark what is extracted from existing projects versus what is speculative invention.
What Good Research Help Looks Like
- Starts from the user’s research taste and long-horizon goals.
- Identifies the boring/incremental version and avoids it.
- Names the strongest possible claim and the experiment that could kill it.
- Preserves top-conference reviewer skepticism.
- Suggests project-level pivots when a local patch would only make a weak idea more polished.
Counterpoints and Gaps
- Radical reframing can waste time if it ignores already completed evidence.
- Novelty should not become vague ambition; every proposed idea still needs a path to proof.
- Some tasks are execution tasks, not ideation tasks; in those cases the agent should respect the existing project contract.