The pandemic was supposed to be EdTech's moment. The great democratiser. Move every lesson online, and geography and privilege stop mattering. The opposite happened. When learning went digital, the children with a quiet room, a dedicated laptop, fast fibre and a parent or paid tutor to fill the gaps pulled further ahead. The children sharing one phone between siblings in a low-cost flat or a rural longhouse simply fell off the map. Technology did not close the gap. It amplified whatever support a child already had at home.

In 1984, Benjamin Bloom published a paper that has haunted education policy ever since. He reported that the average student tutored one-to-one using mastery learning Definition The pedagogical approach of formative testing and corrective feedback until students master each unit before moving on. Bloom paired it with one-to-one delivery to produce the famous 2-sigma effect. performed about two standard deviations better than students in a conventional classroom. The average tutored student landed above ninety-eight per cent of the children in the control group ^[?] . He called it the "2 Sigma Problem", not as a sales pitch for tutoring, but as a challenge: how do we make group instruction as good as one-to-one, since one-to-one for every child is unaffordable?

The modern footnote: most meta-analyses do not reproduce a clean 2-sigma. Nickow, Oreopoulos and Quan's 2020 review of ninety-six randomised tutoring trials puts the pooled effect at about 0.37 standard deviations, roughly fourteen percentile points ^[?] . Impressive, but nowhere near two sigma. Treat 2-sigma as the theoretical ceiling of fully individualised mastery instruction, not the typical result of a crowded after-school tuition centre.

But the ceiling is rising. A 2024 Harvard study of 194 students in the Physical Sciences 2 course found that students using a GPT-4-based AI tutor "learned twice as much content in less time" as students in a conventional class ^[?] . The upside of one-to-one, human or hybrid, is arguably larger in 2026 than when Bloom wrote.

Daily volume is not the variable. Even the highest-performing systems cannot tutor every child. The right question is who needs personalisation most, and where the gap is most catastrophic. Map two axes (school and home support during disruption against access to personalised tutoring) and four Malaysian archetypes emerge.

Three conditions sit on a spectrum from Bloom's theoretical ceiling through current reality to the emerging AI-tutored frontier. The headline effects, by the literature.

Three Malaysian numbers carry the realised gap.

The simplest possible comparison: where Malaysia was, where Malaysia is. Drag the slider to see the post-COVID collapse in a single chart.

The policy point sits in the matrix. The children who would benefit most from a sigma of personalised gain are precisely the ones least able to buy it ^[?] . Mass schooling alone can no longer close this gap. Structured, individualised instruction is shifting from a middle-class luxury to a national necessity, especially for the B40.

Pick one child who fell behind during the closures. Run a real diagnostic this week. Start a single targeted one-to-one session against their actual gaps, not the syllabus, the gaps. That is how you begin to close a two-sigma problem. One student at a time.