Confidence-Weighted Adaptive Half-Life Scheduling (CW-AHS): A Lightweight Online Alternative to Batch-Trained Spaced Repetition Models

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Spaced repetition systems schedule reviews of learned material to maximize long-term retention for a fixed number of practice opportunities. Widely used heuristics such as the Leitner system and SM-2 (the algorithm behind Anki's classic scheduler) adjust review intervals using fixed multipliers that are not grounded in an explicit model of forgetting, while machine-learned alternatives such as half-life regression (HLR) fit an explicit exponential forgetting model but require periodically retraining on batches of accumulated review logs—an engineering burden that is difficult to support in a lightweight, single-user study application. This paper proposes Confidence-Weighted Adaptive Half-Life Scheduling (CW-AHS), an online scheduler that maintains a per-item half-life estimate and updates it after every review using a multiplicative rule driven by the gap between predicted and observed recall, requiring no offline training step. In a controlled simulation of 200 synthetic learners studying 30 items each over a 60-day horizon, CW-AHS matches the review budget of simple heuristics (13.2 reviews/item vs. 13.3–15.2 for SM-2 and Leitner) while achieving higher exam-day retention (71.5% vs. 63.8–70.4%), and attains the best retention-per-review efficiency of all four schedulers tested, including the batch-trained HLR baseline. Source code for the algorithmic simulation is openly available at GitHub: https://github.com/zyeeio/CW-AHS-Spaced-Repetition