One of FSRS's biggest advantages over older algorithms is that its parameters can be refit to your personal review history. Optimization is the process of taking everything you've ever done in the app and asking: given this learner's actual forgetting patterns, what weights would have predicted those outcomes best?
Done right, it can tangibly reduce your review workload for the same retention. Done wrong, it makes your scheduler worse. Here's how to tell the difference.
When Optimization Helps
You have enough review history. The FSRS team's published guidance suggests a minimum of around 1,000 total reviews and at least a few hundred cards before optimization produces meaningfully better parameters than the defaults. Below that threshold, you're fitting noise.
Your material is reasonably consistent. Optimization fits a single set of parameters to your overall forgetting behavior. If you have one deck of medical facts and another of song lyrics, those are very different cognitive tasks, and the fit will be a compromise. Optimization per deck helps, and FSRS supports it.
Your study habits are stable. If you're switching between heavy and light study weeks erratically, the model has trouble separating signal from scheduling noise.
When Optimization Hurts
You're early. Running the optimizer at 100 reviews will give you parameters that look precise but are actually overfit to a tiny sample. You'll feel the impact over the next month as the scheduler mispredicts forgetting on cards you haven't seen yet.
Your review history is dominated by one type of rating. If you press Good on 95% of your reviews (common for easy material), the optimizer has almost no signal to distinguish stability from difficulty.
Your rating habits are inconsistent. FSRS assumes your ratings mean something. If you press Easy when you mean Good, or Hard when you mean Again, the optimizer fits that inconsistency and bakes it in.
How Often Should You Re-Optimize?
For most learners, every 2–3 months is plenty. More often is noise. Less often is fine as long as your study habits haven't changed dramatically.
Does Optimization Rewrite My Existing Schedule?
No. Optimization changes the weights FSRS uses for future predictions. Cards you've already reviewed stay where they are until their next review, at which point the new parameters take over. It's a gradual handoff, not a reset.
A Practical Workflow
- Use the defaults for your first month.
- Cross the ~1,000-review threshold.
- Run the optimizer.
- Review the new parameters — if they look wildly different from the defaults, that's a signal something is unusual about your review pattern (possibly your ratings).
- Give the new parameters two weeks before evaluating.
- Re-optimize every couple of months thereafter.
The Bigger Point
Optimization is a power tool, not a magic button. It rewards learners with enough review data and consistent habits, and it punishes learners who run it too eagerly. Neurako's optimizer is built to be used — but it's also built to refuse to run when there isn't enough data to give you a meaningful result.
Sources
- FSRS4Anki wiki: github.com/open-spaced-repetition/fsrs4anki/wiki
- FSRS benchmark: fsrs-benchmark