ADR-0004: Four-Pillar Pedagogical Assessment Model

Status

Accepted

Context

Traditional flashcard spaced repetition models measure recall as a 1-dimensional score. However, real-world engineering and technical mastery requires multiple cognitive competencies: theoretical comprehension, practical application, debugging proficiency, and communication/teaching ability.

Decision

We implemented a four-pillar pedagogical assessment model in src/engine/mastery.ts:

• Conceptual ($20\mathrm{\%}$ weight): Theoretical grasp of core principles.
• Practical ($20\mathrm{\%}$ weight): Ability to apply concepts in code or execution.
• Debug ($20\mathrm{\%}$ weight): Ability to troubleshoot, identify root causes, and fix failures.
• Feynman ($40\mathrm{\%}$ weight - double weighted): Ability to articulate and explain concepts simply without jargon.

Mastery calculation formula:

topic_mastery = round((conceptual + practical + debug + 2 * feynman) / 5, 4)

A topic is considered mastered and satisfies downstream dependencies when its weighted score reaches or exceeds 0.70 (70%).

Consequences

• Positive:
  • Encourages holistic comprehension rather than rote memorization.
  • Double weighting the Feynman technique emphasizes deep conceptual clarity.
  • Clean floating point normalization clamped to $[0.0,1.0]$.
• Negative / Tradeoffs:
  • Requires evaluating topics across 4 dimensions rather than a single metric.

Alternatives Considered

1. Unweighted Equal Average (25% per Pillar):

   • Description: $\text{mastery}=(c+p+d+f)/4$.
   • Pros: Simpler conceptual formulation.
   • Why Rejected: In accordance with Richard Feynman's learning heuristics, the ability to articulate and explain a concept simply without jargon serves as a primary indicator of genuine comprehension. PALEE deliberately double-weights the Feynman dimension (40%) as an architectural design decision to penalize superficial rote memorization and prioritize clear technical articulation.

2. 1-Dimensional Traditional Recall Score (0.0 to 1.0):

   • Description: Single scalar score representing overall familiarity.
   • Pros: Low cognitive friction when recording a review.
   • Why Rejected: Conflates theoretical memorization with practical execution. A developer might memorize a concept's definition but fail to debug a production outage or implement it in code.

3. Bloom's Revised Taxonomy 6-Tier Scoring Hierarchy:

   • Description: Evaluating Remember, Understand, Apply, Analyze, Evaluate, Create independently.
   • Pros: Academically comprehensive educational framework.
   • Why Rejected: 6 distinct evaluation dimensions impose excessive cognitive overhead on everyday study workflows. Four targeted dimensions (Conceptual, Practical, Debug, Feynman) capture software engineering workflows cleanly.

4. Time-Decayed Dynamic Mastery:

   • Description: Continuously decreasing mastery scores as time passes since the last review.
   • Why Rejected: Conflates retrieval scheduling with fundamental competence. Temporal decay is already handled by SM-2 due_at and ease_factor intervals; modifying the underlying mastery score creates dual-decay instability in dependency graph prerequisites.