Restoring the Credibility of AIGC Rate: An AIGC Rate Calibration Framework Based on Metacognitive Contribution (MS) and Machine Evidence (MR)

Current evaluation dimensions of AIGC detection overly focus on the "suspected AI text rate" while neglecting the contribution differences between humans and AI in effective output. Such tools rely on modeling text statistical features, making them unable to counter adversarial tactics like semantic paraphrasing and cross-linguistic translation—exposing fundamental flaws in the single-feature detection paradigm.This study introduces a metacognitive perspective and proposes the "MS×MR" calibration framework, which takes cognitive contribution (MS) and machine detection evidence (MR) as dual evidence. Through conflict direction- and intensity-aware fusion, the framework outputs trustworthy and auditable suspected AIGC rates.Empirical results show that:On cross-linguistic datasets (M=10, equal-width binning), the Brier score of MR-only decreased from 0.003323 to 0.000459, ECE from 0.03606 to 0.00963, and MCE from 0.0893 to 0.03303.Conclusions remained robust across different binning schemes (M=10/20/50; equal-width/equal-frequency), with improvements in ΔBS/ BSS being significant under paired permutation tests (p<0.001) ； Global 1000×1000 grid analysis reveals that:The proportion of NEG was approximately 56%; the proportions of strong/medium/weak intensity were roughly 29.9%/30.4%/39.7%.The HUMAN_REVIEW (HR) trigger rate was about 10.27%, which was highly consistent with the POS×STRONG combination, forming a quantifiable review budget boundary ； In user interviews (n=60):The persuasiveness of the report increased from 2.03±0.78 to 3.48±0.62.Approximately 98.3% of participants reported reduced outcome anxiety.For samples of "non-malpractic cross-linguistic polishing/translation," HR=0, and the framework balanced both calibration and resolution.Overall, through the dual calibration of "cognition × machine," MS×MR overcomes the key flaws of the single-feature paradigm, providing a trustworthy, robust, and interpretable AIGC rate detection solution for high-stakes scenarios such as academic integrity.