Research · Pedagogy

How this site teaches

Calyx runs on its own learning theory. UTML models learning as a move from entropy (surprise) toward coherence (durable, connected understanding), described by four factors — ΔP, ΔK, ΔΩ, ΔΞ. The predict→reveal→explain-back→mastery loop you move through is that theory running as code, not a metaphor bolted on after.

Read the UTML paper

The loop you move through

Every concept walks the same path. Each step is an instance of a UTML factor, implemented by a real feature on this site (docs/04 §2).

  1. ΔP

    Predict

    You commit an answer before the reveal, so the surprise between guess and truth is measured.

    Predict-then-reveal on every concept
  2. ΔP

    Reveal

    The grounded answer is shown only after you commit — feedback lands on a real prediction error.

    The grounded, cited reveal
  3. ΔK

    Explain back

    You say the idea in your own words; the tutor scores it for coherence (self-explanation / teach-back).

    Explain-back, then the grounded tutor
  4. ΔΞ

    Master

    Progress is gated behind a transfer check — fluent talk with a failed transfer banks nothing.

    The mastery gate (practice quiz + Ward)
  5. ΔΩ

    Stay in flow

    Difficulty is tuned toward ~80% success so you stay challenged but not stalled.

    Adaptive difficulty + affect-aware help
  6. ΔK

    Space it out

    Retrieval is scheduled over time so the coherence you built does not decay.

    Spaced review (FSRS)

The four factors, mapped to features

The factors are not a metaphor here — they are the control signals the engine reads (docs/04 §1).

Factor What it is How the site runs it
ΔP — prediction error The surprise that drives an update: the raw signal before grounding. Forces a prediction before every reveal; sizes items so you are genuinely uncertain.
ΔK — coherence Knowledge becoming connected, consistent, and durable — the kernel forming. Drives explain-back and elaboration, then schedules spaced retrieval so it persists.
ΔΩ — optimizer state The health of the learning process itself — keeping you in the flow channel. Holds ~80% success with adaptive difficulty; detects affect and reduces friction.
ΔΞ — non-collapse The guard against illusory coherence — fluent pattern-matching that is not understanding. Gates mastery behind a high-confidence transfer check (a Ward); refuses to certify it unearned.

Why this is honest, not aspirational

These are the same mechanisms the engine actually uses: a prediction is forced before every reveal, mastery is gated behind a transfer check (a fluent explanation alone banks nothing), and review is scheduled so coherence does not decay. The theory is the rigorous source; this page is the map from it to the buttons you click.

Read the UTML paper

← All research