| name | sci-comparative-positioning |
| description | Use when a manuscript claims novelty, improvement, efficiency, speedup, superiority, reduced cost, better performance, or contribution relative to prior work, baselines, surveys, systems, or benchmarks |
SCI Comparative Positioning
Overview
Audit whether comparison language is legitimate before it enters the paper.
Core principle: no "better/faster/improved" claim until task, data, hardware, metric, and evidence level are comparable.
Trigger Words
Run this skill when text contains or implies:
- faster, better, improves, reduces, efficient, outperforms
- SOTA, first, novel, superior, significant gain
- baseline, comparison, prior work, compared with
- practical value over existing systems
Required Tables
Comparable Literature Matrix
| Work | Task | Hardware/environment | Metric | Evidence level | Comparable? |
|---|
Comparison Validity Table
| Claim type | Required conditions | Allowed wording |
|---|
| Direct quantitative improvement | same task, metric, data, and controlled hardware | "X improved by Y under Z." |
| Weak quantitative reference | similar metric, different task/hardware/data | "same order/context, not direct baseline." |
| Qualitative positioning | different layer or workflow | "Unlike model-level studies, this evaluates an end-to-end workflow." |
| Not comparable | missing or incompatible basis | Do not write improvement language. |
Contribution Delta Table
| Prior layer | Current paper layer | Safe wording |
|---|
Evidence Gap Table
| Desired stronger claim | Missing evidence | Recommended experiment |
|---|
Output
- safe comparative paragraph
- prohibited phrases
- evidence gaps
- recommended experiments
Red Flags
- Comparing model latency to full workflow latency.
- Comparing different datasets as if controlled.
- Using survey claims as baselines.
- Turning COTS HIL evidence into flight validation.
See references/pressure-scenarios.md for reusable failure cases from satellite/edge-agent manuscript practice.
