Q001-SRC02-E01¶


Research	R0049 — Landscape Scan
Run	2026-03-31
Query	Q001
Source	SRC02
Evidence	E01

Extract¶

Framework Chain-of-Thought (FCoT) is a novel prompting approach that directs LLMs to systematically reason against predefined frameworks when screening papers for systematic reviews. Evaluated across ten systematic reviews covering four common types of SR questions, FCoT achieved a mean accuracy of 93.6% and sensitivity of 97.5% in full-text screening. The approach encodes review inclusion/exclusion criteria as structured reasoning steps for the LLM.

Relevance to Hypotheses¶

Hypothesis	Relationship	Strength
H1	Partial support — demonstrates that framework-guided prompting works but covers only screening	Moderate
H2	Contradicts — this is a published prompt implementing a research methodology component	Strong
H3	Supports — exemplifies effective but phase-limited implementation	Strong

Context¶

FCoT is the most methodologically rigorous prompt design found during this research. It demonstrates that encoding analytical frameworks into LLM prompts produces measurably better results than unconstrained prompting. However, it addresses only the screening phase of systematic reviews — not evidence scoring, synthesis, confidence calibration, or self-audit.

Notes¶

The 93.6% accuracy figure is significant because it approaches human performance levels for this specific task. The principle of framework-guided prompting (rather than open-ended instructions) is transferable to other research phases, suggesting that the approach could be extended to build a more comprehensive framework.