What we build on, graded honestly.

Retrieval practice (the testing effect) shows a pooled advantage of roughly g = 0.50 over restudying, rising to about 0.73 when the recall comes with feedback (Rowland, 2014). The honest caveat: most trials are educational, with few run inside corporate learning.

Shapes: every Stica Apply prompt asks you to produce, never to re-read

Distributing practice over time improves retention, around g = 0.28 in applied settings (larger in the laboratory). It only works if the learner actually comes back, which is why the schedule in Stica Apply runs itself instead of waiting for willpower.

Shapes: the fixed, system-run arc in Stica Apply

Unused skills fade measurably within a month of training and keep fading, which is why the first real application is designed into the plan's opening weeks rather than left to "someday". Decay curves for interpersonal skills are less well mapped, and we say so.

Shapes: the early "use it for real" step in every plan

Implementation intentions ("when X happens, I will do Y") reliably help people start a behaviour. The classic large effect has shrunk in the most recent pooled analysis (2024, over 600 tests); the technique still earns its place, but for initiation more than maintenance, and we use it that way.

Shapes: plans anchor the first application to your own named situation

The established diagnostic frameworks (COM-B and its relatives, built from a synthesis of nineteen behaviour-change frameworks) separate capability from opportunity and motivation. The practitioner triage of "can't do" versus "won't do" has thin trial evidence but decades of field use; we apply it, and we tell you when your problem looks like environment or motivation rather than a course.

Shapes: the diagnosis stage of every Stica Craft analysis, and its willingness to say "this is not a training need"

Transfer climate, whether the workplace supports the new behaviour, explains a large share of whether training transfers (climate R² = .32 in the strongest pooled analysis; supervisor support correlates at roughly r = .31). The main transfer meta-analysis is now dated and rests on self-report, so we keep the claim modest, but the direction is one of the most consistent findings in the field.

Shapes: plans engage a manager, a peer or a default, not solo willpower

Wanting to use what you learned is among the strongest predictors that you will (motivation to transfer, roughly r = .37, same dated-anchor caveat as above). Relevance framing helps too: showing people the personal usefulness of material has a pooled effect around d = 0.49, though much of that evidence comes from education rather than the workplace.

Shapes: plans built from your own goal and material, not a generic syllabus

Specific, self-authored goals are one of the strongest individual levers in the literature, with the caveat that tight goals can harm novel, complex tasks. It is why the personal door of Stica Craft drafts a plan from your own answers and reflection, rather than assigning you one.

Shapes: the guided capture and reflection questions in every personal plan

Habits typically take two to five months to form, with a documented range from days to nearly a year. The popular "66 days" figure is a misread mean from a single small study and we never use it as a target. Our arcs end when the behaviour holds, not on a fixed date.

Once publication bias is corrected, the average nudge effect is close to zero (around d = 0.04 in the most critical pooled analysis). We use prompts and defaults as amplifiers inside a designed programme, never as the intervention itself.

Adaptive and tutoring systems show real pooled gains on knowledge outcomes, but there is no pooled trial base yet for the newest AI tutors, engagement with digital behaviour-change tools decays without human structure, and AI assistance can improve output while leaving retention flat unless recall is designed in. Stica's answer is exactly that design: the AI drafts and schedules, the practice makes you produce, and we treat claims beyond knowledge acquisition as hypotheses to measure.