What a scheduler cannot know

Judgement

A scheduler knows only what it is told. In the standard FSRS optimiser schema, the required history is essentially which card was reviewed, when it was reviewed and the learner's subjective rating. Review duration and card state can be optional. The schema does not include the card text, answer text, conceptual importance, exam date, workplace consequences or real-world performance.¹

The rating is not an independent observation of memory. The learner chooses a category describing how recall felt, and the scheduler updates from that category. Reference guidance treats forgetting differently from a difficult but successful recall. If the learner reports a failure as a difficult success, the model receives misleading history.² Anki therefore warns that its Hard rating is a passing grade and should not replace the failure rating after forgetting.³

Priorities also come from outside the model. Desired retention, learning steps and maximum interval are configuration choices supplied to a reference FSRS scheduler. Its documented constructor has no item-importance or exam-deadline input.⁴ The scheduler can calculate an interval consistent with its settings, but it cannot decide how costly forgetting would be for this learner. A more important target must be expressed through deck design, settings or an application layer.

With a deck and twenty minutes, rate recall honestly and inspect what the schedule cannot inspect. Ask whether the card is clear, whether the answer is understood, whether similar items are interfering and whether the knowledge still matters. A due card is not automatically an important card. A confident response to familiar wording is not automatically success on a differently phrased question or real task. The model's recall prediction concerns the card presented to it, not transfer.²

Richer applications can provide more information. Duolingo's published model, for example, used linguistic features alongside learner-performance data.⁵ Additional fields can change what a model receives, but a standard review log does not directly supply priorities or circumstances that were not recorded.

Boundary

Log-aware scheduling is a reasonable procedure built on strong spacing and retrieval findings. Its boundary is the information supplied. Understanding, cue quality, sleep, interference and motivation are invisible to a standard review log, and no study in this dossier quantifies how much those omissions cost. Scheduler precision about same-card recall must not be treated as precision about understanding or performance elsewhere.