Markdown version: https://ai.hiina.space/reports/jp-learning-and-anki-tooling/03-scheduler-settings-fsrs-6-on-anki-26.md ยท index

Index of everything on this site: https://ai.hiina.space/index.md

Project status: Japanese immersion learning with Anki and selective glossing: 3. Scheduler settings (FSRS-6 on Anki 26.x)

3. Scheduler settings (FSRS-6 on Anki 26.x)

Anki is on a 26.x release train. 26.08.1 is the latest stable release listed, 26.09 betas ship FSRS 6.6.2, and FSRS-6 (21 parameters) is the current scheduler (Anki releases; awesome-fsrs: The Algorithm). The years on the release page are inferred from ordering. 26.08 also added an "Experiments" section, including an editor rework that "may affect addon compatibility" (Anki 26.08). Since Yomitan, asbplayer and any agent tooling all go through add-ons, the plan leaves those experiments off (inference).

Planned values, sourced from the Anki manual and the vnclub Japanese guide, the two primary sources that loaded:

Setting Value Source / reason
FSRS On Current scheduler
Desired retention 0.90 (manual default); 0.85 if daily load is uncomfortable; stay below 0.97 Workload climbs steeply above 0.90 (Deck Options; vnclub Anki guide). "Help Me Decide" shows the workload curve; it replaced the minimum-retention tool in 25.07
Optimise Monthly, with "Check health when optimizing" on The health check flags pressing Hard on a forgotten card as a data problem
Grading Again and Good only, via the PassFail2 add-on (876946123) Set "Ignore cards reviewed before" to the date pass/fail grading started if earlier history used Hard/Easy
Steps One 10-minute learning step; about 3 minutes relearning vnclub; TheMoeWay mining
Limits Max reviews/day 9999; next day starts at 4 AM Same
New cards 10-15 a day, set to 0 when reviews pile up Cap by load, not a fixed number (vnclub)
Load balancing, Easy Days Core features; 26.05 improved load balancing Older advice to install FSRS Helper for these is probably dated (inference; the Helper repo did not load)
Avoid "Reschedule cards on change" without a backup; the AutoReorder add-on Lapis's README warns AutoReorder has sometimes reset learned cards to new (Lapis README)

The plan keeps one preset unless the health check says otherwise, because the manual reserves separate presets for decks of very different difficulty. One inference to test: newer Anki allows per-deck retention within a preset, so Lapis::<show> premine decks could run at 0.85 while that show is being watched, since the episodes re-expose those words for free. No primary source makes that argument.


Text licensed CC BY 4.0 (attribution: Hiina, ai.hiina.space). Source: https://github.com/hiinaspace/ai-hiina-space