CogEvolution's simulated learning trajectory fits the human power-law trajectory far better than baselines (R²LC 0.92 vs 0.78 PEERS, 0.45 static agents)¶

Claim · All claims
Evidence · 1 study (3 entries) · design r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? CogEvolution produces an error-rate trajectory that tightly aligns with the ground-truth human trajectory (R²LC = 0.92), with a steep drop in the first 20 steps followed by gradual stabilization. → Wei Zhang 2026 q2 i? Static agents exhibit a nearly flat trajectory (R²LC = 0.45), showing no learning effect from practice without a state update mechanism. → Wei Zhang 2026 (2) q2 i? The PEERS model shows a downward but overly linear trend (R²LC = 0.78), struggling to capture diminishing returns characteristic of cognitive saturation. → Wei Zhang 2026 (3)

Evidence¶

Wei Zhang 2026¶

Wei Zhang, Yihang Cheng, Zhirong Ye & Kezhen Huang. (2026). CogEvolution: A Human-like Generative Educational Agent to Simulate Student's Cognitive Evolution. https://arxiv.org/abs/2604.14786 doi:10.48550/arxiv.2604.14786

q2 · i? · design · r2

Learning-curve fitting analysis for RQ2 over 100 practice opportunities on CogMath-948 (Figure 2) fits each agent's error-rate sequence to the power function E(n)=A·n^-α; CogEvolution reaches "R²LC =0.92" against the ground-truth human trajectory. No effect size is printed.

"In sharp contrast,CogEvolution(Blue Circle) produces a curvethattightlyalignswiththeGroundTruth(𝑅 2 𝐿𝐶 =0.92). The trajectory displays a steep drop in the first 20 steps, followed by a gradual stabilization."

Wei Zhang 2026 (2)¶

Wei Zhang, Yihang Cheng, Zhirong Ye & Kezhen Huang. (2026). CogEvolution: A Human-like Generative Educational Agent to Simulate Student's Cognitive Evolution. https://arxiv.org/abs/2604.14786

q2 · i? · design · r2

The same Figure 2 learning-curve analysis reports static agents' "nearly flat trajectory (𝑅2𝐿𝐶 =0.45)", which the authors interpret as standard LLMs relying on pre-trained knowledge without manifesting learning from practice.

"As observed in Figure 2, theStatic Agents(Grey Square) exhibit a nearly flat trajectory (𝑅2𝐿𝐶 =0.45)."

Wei Zhang 2026 (3)¶

Wei Zhang, Yihang Cheng, Zhirong Ye & Kezhen Huang. (2026). CogEvolution: A Human-like Generative Educational Agent to Simulate Student's Cognitive Evolution. https://arxiv.org/abs/2604.14786

q2 · i? · design · r2

In the same RQ2 analysis, PEERS achieves "R²LC =0.78" with an overly linear trajectory; the authors attribute this to BKT-based probability transitions failing to capture diminishing returns of cognitive saturation.

"ThePEERSmodel (Green Triangle) shows a downward trend (𝑅2 𝐿𝐶 =0.78), but its trajectory is overly linear. This suggests that while BKT-based probability transitions can simulate performance improvement, they struggle to capture thediminishing returns characteristic of cognitive saturation."

Discussion¶

Related Claims¶

• Ablations show each CogEvolution module contributes: removing ICAP perception, structured retrieval, or evolutionary update degrades mistake precision, learning-curve fit, and alignment — related
• CogEvolution performs comparably to the KT-based PEERS model on mastery prediction (AUC 0.80 vs 0.82) while achieving higher mistake precision (76.8%) on CogMath-948 — related
• Automatized L2 morphosyntactic skills develop very gradually over time, following a power function learning curve like other cognitive skills — related