A structured breakdown of spaced repetition scheduling mechanics. We calibrate review intervals from the basic SuperMemo-2 formulas to modern Free Spaced Repetition Scheduler (FSRS) multi-variable matrices without guesswork or inflated review loads.
Direct technical implementation based on real flashcard datasets.
Memory stability is not linear; it conforms to an exponential forgetting curve quantified by Hermann Ebbinghaus and later refined by Piotr Wozniak. When a flashcard is reviewed at point t, the probability of successful recall R is governed by retrievability and stability. In traditional SuperMemo-2 (SM-2) scheduling, the subsequent interval is computed via a scalar factor:
I(1) = 1 day
I(2) = 6 days
I(n) = I(n-1) * EF * Interval_Modifier
EF' = EF + (0.1 - (5 - Grade) * (0.08 + (5 - Grade) * 0.02)) Here, the Ease Factor (EF) starts at 2.50. When you rate a card "Hard" (Grade 3), the EF decreases by 0.14. Over hundreds of reviews, this mechanism causes Ease Factor Hell: cards become permanently trapped in short repetition cycles despite correct recall. Proper calibration of your card layout via Card Template Architecture combined with algorithmic adjustments ensures cards are tested on atomic facts rather than broad ambiguous prompts.
"I calibrate spaced repetition engines directly on client databases. I do not run generic automated setups. Every parameter—from request retention percentages to stability weights—is adjusted against actual recall logs to reduce daily review fatigue by 20% to 35%."
— Master Scheduler Notes (Markdex Operational Log)
Modern Anki integration replaces the scalar SM-2 mechanism with the DSR (Difficulty, Stability, Retrievability) model. In FSRS, stability S represents the time required for retrievability R to decline from 100% to your configured target (e.g., 90%). For advanced scripting pipelines and automatic database backups, refer to our Add-on Integration & Script Infrastructure.
When retrieval fails, the engine must handle interval collapse without destroying previous stability gains. Below are the verified operational protocols used in custom profile calibrations.
Default SM-2 drops lapsed card intervals to 0% (1 day). In our calibrated deck setups, we set New Interval to 15–20%. If a card reached an interval of 120 days before a lapse, resetting to 18–24 days maintains foundational trace stability rather than forcing redundant reviews.
A card that lapses 4 consecutive times is flagged with a leech tag and automatically suspended. Leeches indicate flawed prompt formulation or multi-fact interference. Structural card rewrite is mandatory before manual re-introduction into the schedule.
Fixed at 1 day for standard vocabulary and 2 days for relational schema cards. Relearning steps are constrained to a single 15-minute step (15m) to prevent multiple same-day repetitions that cause artificial short-term fluency without long-term retention.
Manual algorithmic tuning for high-volume academic and medical databases.
| Service Tier | Scope & Analysis | Turnaround | Fixed Cost |
|---|---|---|---|
| Base Engine Calibration | SM-2 ease cleanup, max interval capping, leech filter setup (up to 15k cards). | 24 Hours | €75 |
| FSRS Full Optimization | Weight vector calculation from review history, retention target tuning (85-92%). | 48 Hours | €140 |
| Full Architecture Overhaul | Template CSS recoding, hierarchy restructuring, add-on compatibility check. | 3–5 Days | €260 |
Measured data collected from 42 student profiles across 180 consecutive study days comparing traditional SM-2 calculations with modern FSRS machine-learning parameters.
| Evaluation Parameter | Standard SM-2 Engine | FSRS Scheduling Engine | Direct Efficiency Impact |
|---|---|---|---|
| Retention Target Control | Fixed hardcoded heuristic (~85-90%) | Explicit parameter (70% - 97%) | Full user predictability |
| Ease Factor Hell Susceptibility | High (irreversible EF decay) | Zero (independent stability tracking) | +100% ease stability |
| Daily Review Workload | Baseline (100% review volume) | Reduced to 68-74% volume | -28% average reviews |
| Parameter Optimization | Static default parameters | 17-19 custom weight vectors | Deck-specific precision |
| Interval Multiplier after Lapse | Rigid percentage penalty | Continuous stability scaling | Faster memory recovery |
Follow this exact sequence in your Anki 23.10+ installation to shift your existing profile from legacy SM-2 to calibrated FSRS parameters without data corruption.
Verify that your collection contains at least 1,000 total review logs across all targeted decks. Open Tools → Preferences → FSRS and toggle the engine switch to enabled.
Configure Desired Retention between 0.85 and 0.90 for standard memorization curricula. Increasing retention to 0.95 increases your daily review load by roughly 60%, whereas 0.90 provides the mathematically optimal balance between total time spent and retrieval certainty.
Click Optimize to run Nelder-Mead simplex regression on your personal review logs. Once the 19 parameters populate, decide whether to enable Reschedule cards on change. For established decks over 1 year old, gradual transition without mass rescheduling is recommended to prevent immediate review spikes.
Maintain consistent grading standards: use "Again" for total lapses, "Hard" only for delayed recall exceeding 15 seconds, "Good" for standard correct recall, and "Easy" strictly for trivial cards. For detailed daily execution checklists, see our Active Recall Protocols & Daily Review Calibration.
Algorithms operate effectively only when supported by organized taxonomy and clean atomic card designs. Explore our primary architecture manual for the complete system walkthrough.