Show & Tell — Phase 27.4 CurriculumDesigner: Adaptive Learning Curricula

[web3guru888]

CurriculumDesigner Architecture

The Problem: Learning Order Matters

Random presentation of learning material is suboptimal. Bengio et al. (2009) showed that training on examples ordered from easy to hard significantly improves generalization. CurriculumDesigner automates this insight — designing and adapting learning curricula in real time.

Zone of Proximal Development

    Difficulty ──────────────────────────────────────▶

    ┌─────────────┬──────────────────┬───────────────┐
    │  MASTERED   │       ZPD        │  TOO HARD     │
    │             │  ╔═══════════╗   │               │
    │  Already    │  ║ LEARNING  ║   │  Prerequisites│
    │  known      │  ║ HAPPENS   ║   │  not met      │
    │             │  ║ HERE      ║   │               │
    │             │  ╚═══════════╝   │               │
    └─────────────┴──────────────────┴───────────────┘
         mastery ≥ 0.8        readiness ≥ 0.8
                              mastery < 0.8

CurriculumDesigner targets the Zone of Proximal Development (Vygotsky, 1978) — objectives that are achievable with effort, neither too easy nor too hard.

Prerequisite DAG

# Example prerequisite graph
objectives = [
    LearningObjective("algebra", prerequisites=frozenset()),
    LearningObjective("calculus", prerequisites=frozenset({"algebra"})),
    LearningObjective("linear_algebra", prerequisites=frozenset({"algebra"})),
    LearningObjective("optimization", prerequisites=frozenset({"calculus", "linear_algebra"})),
    LearningObjective("ml_basics", prerequisites=frozenset({"optimization"})),
]

# Topological sort → ["algebra", "calculus"|"linear_algebra", "optimization", "ml_basics"]
# With ZPD filtering → next = first objective where readiness ≥ 0.8 and mastery < 0.8

algebra ──▶ calculus ──────────▶ optimization ──▶ ml_basics
   │                                 ▲
   └──▶ linear_algebra ─────────────┘

Adaptive Difficulty

async def adapt_difficulty(self, curriculum, recent_performance):
    avg = sum(recent_performance) / len(recent_performance)
    if avg > 0.9:
        # Too easy — skip ahead or increase difficulty
        new_rate = min(curriculum.adaptation_rate * 1.5, 1.0)
    elif avg < 0.5:
        # Too hard — revisit prerequisites or slow down
        new_rate = max(curriculum.adaptation_rate * 0.5, 0.01)
    else:
        new_rate = curriculum.adaptation_rate  # In the ZPD — stay the course
    
    return Curriculum(
        curriculum_id=curriculum.curriculum_id,
        objectives_ordered=curriculum.objectives_ordered,
        current_position=curriculum.current_position,
        mastery_scores=curriculum.mastery_scores,
        adaptation_rate=new_rate,
        mastery_threshold=curriculum.mastery_threshold,
    )

Plateau Detection

When mastery stops improving despite practice, the system detects a learning plateau:

Signal	Detection	Response
Score stagnation	Δmastery < 0.01 for 5 assessments	Flag plateau, consider different approach
Oscillation	Score alternates ±0.1	Increase support examples, revisit prerequisites
Ceiling effect	Score ≈ 0.8 but never reaches threshold	Lower threshold slightly or provide scaffolding

Integration

GoalDecomposer (10.2)        FewShotAdapter (27.3)
  decomposed goals    ──▶      adaptation results
       │                            │
  LearningObjectives         mastery assessment
       │                            │
       └────────▶ CurriculumDesigner (27.4) ◀────────┘
                        │                │
                  ReflectionCycle     CuriosityModule
                    (16.5)              (13.3)
                  progress review    exploration override

---

"Start small" is not just developmental advice — it's an optimal learning strategy (Elman, 1993).
Issue: #597 | Planning: #593