psitta

Documentation Index

Hypotheses and Scope of the SRS Model

This document describes the non-mathematical hypotheses of the spaced repetition model (SRS) used in the application.

It complements:

Its purpose is to make explicit the cognitive, pedagogical, and product choices that guide the model, in order to:


1. Core Hypothesis: the SRS evaluates a task, not abstract knowledge

Each exercise (in the sense of: type of task) has its own SRS state.

The same content (word, rule, sentence) may therefore be associated with several distinct exercises, for example:

The SRS never evaluates a “global mastery” of a word or rule, but only the ability to succeed at one specific task.

This choice allows:


2. Hypothesis on the response signal: subjective self-assessment

The model assumes that the user is capable of honestly self-assessing after each exercise.

The possible responses (Easy / Good / Medium / Hard / Again) reflect:

Actual measured time is not used. The signal is intentionally subjective.

This choice is deliberate because:


3. Hypothesis on delay: success takes precedence over elapsed time

A successful response to an exercise is never penalised, even after a significant delay.

The model considers that:

Delay is only taken into account when the exercise is failed, in order to:

This choice favours:


4. Hypothesis of local independence between exercises

Each exercise is treated as independent from the others.

The SRS does not model:

This choice is intentional.

It rests on the idea that:

The approximation is considered acceptable as long as:


5. Hypothesis on the role of sentences

Sentences are not fundamental units of knowledge.

They serve primarily to:

The SRS for sentences may be less precise than that for words without compromising overall learning.

Aids (e.g. displaying word translations within a sentence) are acceptable and do not invalidate the exercise, since the primary objective remains exposure and comprehension.


6. Product hypothesis: priority on simplicity and explainability

The model favours:

It does not aim for maximum theoretical optimality.

This implies in particular:

These limitations are accepted for the MVP.


7. Session management (MVP)

Sessions are ephemeral objects.

Their role is to:

Exercise prioritisation (e.g. if 150 exercises are due but the daily maximum is 100) is performed before the session, in the Application layer.

Inside a session:


8. Cognitive load management (MVP)

The model has no global representation of the user’s state (fatigue, declining performance).

For the MVP, management relies on simple rules:

Occasional errors or a poor session should not permanently penalise the SRS state.


9. Explicit MVP scope

The MVP does not attempt to solve the following problems:

These evolutions are considered post-MVP and must not influence current choices as long as the hypotheses above are respected.


10. Role of this document

This document serves as a reference for:

Any major SRS modification must be evaluated against the hypotheses described here.


TODO: