Learn → test → retain: a loop grounded in retrieval, spacing, and self-regulation

1. Learn: beyond familiarity

Learning, in the sense that predicts transfer, means you can generate explanations, procedures, or classifications in novel instantiations — not that the page looks recognisable. Familiarity after re-reading is a known trap; retrieval forces a sharper estimate of what you can produce (Roediger & Karpicke, 2006). Early in a topic, intrinsic load is high; use segmentation and concrete examples (Sweller, 1988) before you demand speed.

2. Test: retrieval as the engine of memory change

Tests are not only measurement instruments; in the lab, retrieval attempts modify memory traces. That is why low-stakes quizzing is a pedagogical lever, not punishment (Karpicke & Roediger, 2008). For self-study, “test” means any forced production: blanks, past-paper fragments, explaining without notes, sorting problems by method. The information you want is what broke — that becomes the next learning target.

3. Retain: spacing and the calendar as hypotheses about forgetting

Distributed practice improves long-term retention relative to massing in a wide range of verbal tasks; optimal spacing scales with how long you need to remember (Cepeda et al., 2006). You will not hand-optimise every interval during term — but you can obey the qualitative rule: return before you are completely fluent, increase spacing as performance stabilises, and tighten spacing after errors. Interleaving problem types after initial mastery can further sharpen discrimination (Rohrer & Taylor, 2007).

4. Running the loop when the term is nonlinear

• → Every week includes at least one honest retrieval event per major topic — not only re-reading.
• → Errors drive the next micro-session; do not “cover” new chapters until the last failure mode is named.
• → Spacing is approximate goodness, not perfection; missed days get a shorter, focused return, not a guilt spiral.
• → When the calendar lies, fix the calendar before you fix your character.

What Offload aims to do (without revealing stack or models)

Offload’s product thesis is that students should not need to be part-time scheduling engineers to benefit from retrieval and spacing. We aim to hold the map — what is due, what slipped, what comes next on the calendar — so the cognitive resources freed align with what Dunlosky et al. (2013) already tell us works: practice testing, distributed practice, and elaboration where appropriate. We will not publish our scheduling stack, feature weights, or roadmap of models here; we will state the scientific obligation plainly: if the app does not make honest loops easier in real weeks, it has failed its premise.