Heuristic Learning Systems
apto“I adapt”
Adaptive learning that can prove a student actually learned, not merely that they finished.
Completion is the easiest thing in education to measure and the least worth knowing. apto is built the other way round: every decision it makes about a learner is recorded, attributable, and open to challenge. The claim it makes is not that a course was finished but that a capability was demonstrated, and it can show its working.
Today’s tools measure the wrong thing
Education measures test completion, not genuine learning, and ungoverned AI tutoring is widening the gap, not closing it.
of sub-Saharan students can’t read a basic text by age 10
UN, 2025
children out of school worldwide
UNESCO, 2024
scored on an exam, then the same items 10.5 months later
Biomedical-knowledge retest, PMC10939629, 2024
A test reports completion, not learning
A high score today says little about whether a student understood it, retained it, or can transfer it. The competencies that separate learned from crammed are invisible to any single test.
Ungoverned AI is widening the gap
Heavy AI-crutch users scored about a full letter grade lower on surprise tests 45 days later. Call it “metacognitive laziness”: the work gets done, the learning does not.
Barcaui, 2025 · Fan et al., 2024
How a learner moves through it
Eight pedagogical layers and twenty agents, arranged as a pipeline rather than a catalogue. Each stage hands the next one something more specific than it received.
- 8
- Pedagogical layers
- intake through assessment, every one governed
- 20
- Agents
- fourteen working the pipeline and six governance gates, each with a single defined remit
- 1,000
- Foundational atoms
- the smallest units of knowledge the system reasons over
Intake
Understand the conditions first
A five-phase capture of the learner’s real context: device, connectivity, environment. A platform that assumes a laptop and fibre will fail the learners who need it most, so apto establishes what is actually available before it designs anything.
Objectives
Make the goal specific
Learning objectives are enriched with taxonomy and effort metadata, then ranked. Objectives that score too high on effort are decomposed rather than assigned: an objective a learner cannot finish is not a goal, it is a wall.
Sequence
Build the path
One agent selects the content units; another assigns the modality best suited to each. The sequence is derived from the learner’s profile and objectives, not from a fixed course order.
Delivery
Teach in the right form
Twenty-five modality views across eight categories: video with transcript-grounded tutoring, generated reading notes, diagrams, worked examples, case studies, role-play, simulation. An embedded tutor adapts to the page the learner is on and the objective they are working toward.
Completion is easy to count. That is why it gets counted.
Whether anything was learned is harder, slower, and the only thing anyone actually wanted to know.
Governance that can say no
Most platforms add governance as review: a report produced after the decision. apto runs six gates that a decision must pass before it reaches a learner, and they are non-compensatory: strength in one cannot offset failure in another. An ethics gate can veto outright.
A human oversight layer sits across the pipeline, and learners own their own data. Every grading run logs its ontology hashes and prompt hashes, so any result can be traced back and reproduced, including by someone trying to prove it wrong.
How the gates work- 01Critical inquiry and grounding
- 02Architectural integrity
- 03Vision
- 04Grounding and application
- 05Evaluative governance
- 06Ethics, with the power to veto
Measured, not asserted
The grader is held to a deploy bar rather than a demo. These are the numbers it has to clear.
0.917
Deployed grader agreement
within one band, against a 0.85 deploy bar
3
Independent labs
gold labels cross-validated, not self-scored
1,300+
Academic sources
behind an examined doctoral thesis
The research behind apto is an examined doctoral thesis at the Da Vinci Institute. Its supporting corpus is published on this site as it is released.
Built for institutions that will be asked to justify it
If you will one day have to explain to a regulator, a parent, or a student why the system reached the conclusion it did, the time to require an answer is before deployment, not after.
Why we build
“A girl empowered by the strength of her own voice: a learner profile that speaks for her true abilities; an adaptive system guided by the spirit of equity, that speaks the language of the land, and does not abandon her in struggle.”
Deeply personal, profoundly respectful of culture, resolutely equitable, and forever committed to improvement.
G.D. Hoffman, doctoral thesis, Da Vinci Institute