AnkiWeb addon 215758055
Reorders Anki review cards by expected long-term knowledge gain, using FSRS-based estimates and displaying each card's gain to prioritize the most beneficial reviews.
Open on AnkiWeb GitHub Forum topic Find alternatives
stale
| Min Anki | Max Anki | Updated |
|---|---|---|
| 25.02 | 25.09.2+ | 2025-12-28 |
Loading…
Implements an adaptive FSRS v5/v6 retention scheduler that minimizes expected total review workload until cards reach long-term retirement, though the addon is now deprecated.
Reschedules Anki cards using FSRS review histories, postpones or advances selected cards, balances daily workload, schedules breaks, disperses sibling cards, flattens backlogs, and reports memory and learning-step statistics.
Displays a stacked bar graph of card states over time, color-coded by interval, with an FSRS-based estimated total knowledge retention line and main-screen integration.
Reorders due cloze cards in ascending cloze ordinal order by adjusting their due dates, so reviews follow note sequence; creates backups and is intended for the v3 scheduler.
Generates an HTML dashboard summarizing daily, weekly, or monthly study sessions with retention, FSRS metrics, SVG growth graphs, narrative insights, and optional LLM-generated reflections from local collection data.
Predicts remaining study time and finish time using an offline machine-learning model, with customizable progress displays, metrics, CSS styling, and fatigue alerts.
No scheduler. No leech concerns. No review history burden.
This addon reorders your Anki queue by expected knowledge gain to maximize learning efficiency.
This addon introduces a new review strategy that goes beyond Anki’s built-in options (like "easy cards first" or "descending retrievability"). It prioritizes cards that contribute the most to your long-term memory so that the cards with the highest expected gain are reviewed first.
The long-term knowledge is estimated using an exponential moving average (EMA) of retrievability. This produces a score between 0 and 1 that reflects how well a card is expected to be remembered over time.
Install from AnkiWeb.
215758055.If you encounter any issues or bugs, please open an issue on GitHub.
No scheduler. No leech concerns. No review history burden.
This addon reorders your Anki queue by expected knowledge gain to maximize learning efficiency.
This addon introduces a new review strategy that goes beyond Anki’s built-in options (like "easy cards first" or "descending retrievability"). It prioritizes cards that contribute the most to your long-term memory so that the cards with the highest expected gain are reviewed first.
The long-term knowledge is estimated using discounted retrievability. This produces a score between 0 and 1 that reflects how well a card is expected to be remembered over time.
The future estimator uses a few steps of FSRS simulation to predict the knowledge gain from future reviews.
Install from AnkiWeb.
215758055.The evaluation is based on review-sort-order-comparison, but using an unweighted setting where each review takes equal time. Future versions of this addon may incorporate actual review time.
By default, this addon uses the discounted knowledge (knowledge_gain_discounted_desc). In the experiments, the exam mode (knowledge_gain_delayed_desc) achieved the best performance.
| order | total_learned | total_time | total_remembered | average_true_retention | seconds_per_remembered_card |
|---|---|---|---|---|---|
| knowledge_gain_delayed_desc | 20000 | 96464.0 | 16192 | 0.751 | 5.96 |
| knowledge_gain_discounted_desc | 20000 | 96423.0 | 16030 | 0.728 | 6.02 |
| difficulty_asc | 20000 | 96519.0 | 15737 | 0.793 | 6.13 |
| PSG_desc | 20000 | 96490.0 | 15701 | 0.784 | 6.15 |
| due_date_asc | 20000 | 96508.0 | 15619 | 0.678 | 6.18 |
| random | 20000 | 96483.0 | 15470 | 0.700 | 6.24 |
| retrievability_asc | 20000 | 96512.0 | 15141 | 0.715 | 6.37 |
| stability_desc | 20000 | 96487.0 | 15049 | 0.792 | 6.41 |
| retrievability_desc | 20000 | 96473.0 | 14926 | 0.797 | 6.46 |
| add_order_desc | 20000 | 96508.0 | 14902 | 0.793 | 6.48 |
| PRL_desc | 20000 | 96469.0 | 14383 | 0.793 | 6.71 |
| interval_asc | 20000 | 96470.0 | 14338 | 0.792 | 6.73 |
| stability_asc | 20000 | 96424.0 | 14325 | 0.793 | 6.73 |
| add_order_asc | 20000 | 96501.0 | 13611 | 0.773 | 7.09 |
| interval_desc | 20000 | 96474.0 | 13549 | 0.778 | 7.12 |
| difficulty_desc | 20000 | 96531.0 | 13172 | 0.789 | 7.33 |
This addon estimates long-term knowledge using discounted retrievability:
$$ J_{\text{dis}}(\text{card}, T; \gamma) = -\log \gamma \int_{0}^{\infty} R(\text{card}, T + t) \gamma^t \mathrm{d}t $$
where
For FSRS 4.5 and 5, there’s a closed-form expression for it:
$$ J(\text{card}, T; \gamma) = \sqrt{\pi\alpha\log\gamma} \cdot \text{erfcx}\left(\sqrt{(\alpha-T)\log\gamma}\right) $$
where
For FSRS 6, the discounted knowledge is given by:
$$ J_{\text{dis}}(\text{card}, T; \gamma) = \gamma^{\alpha-T} (\alpha\log\gamma)^{-D} \cdot \Gamma(D+1, (\alpha - T)\log\gamma) $$
where