100 v1.26 Specialist Domains Documented

Knowledge Patterns

Domain-specific pattern detection extracting knowledge-typed patterns from observations — conceptual pattern clustering beyond workflow sequence n-grams.

Structured Data

Capability Metadata

Field	Value
ID	CAP-100
Version	v1.27
Domain	Specialist Domains
Group	Knowledge Packs
Status	Documented
Related	CAP-010, CAP-101, CAP-104
Pattern types	concept-usage, procedure-sequence, error-correction, question-answer
Extends	CAP-010 n-gram pattern detection

Configuration Reference

Knowledge pattern types:

concept-usage — user references same concept multiple times

procedure-sequence — user repeatedly follows same domain procedure

error-correction — user makes same type of domain-specific error

question-answer — user asks the same question across sessions

Pattern confidence decay: patterns expire if not observed for 30 days

Pattern expiry by domain version: patterns invalidated when knowledge pack version increments

Knowledge Pattern Types
────────────────────────
concept-usage:
  User references AGC instruction CCS multiple times
  → suggest adding CCS to knowledge pack exercises

procedure-sequence:
  User repeatedly assembles → simulates → debugs
  → suggest capturing as AGC development workflow skill

error-correction:
  User makes same type error (off-by-one in 15-bit arithmetic)
  → suggest adding common pitfall to knowledge pack concepts

question-answer:
  User asks about copper list encoding 3+ times
  → suggest adding copper list tutorial to knowledge pack

Technical Specifications

Extends: core pattern detection (CAP-010) with knowledge-typed pattern classes

Input: knowledge pack observations from event bridge (CAP-099)

Output: typed pattern records consumed by knowledge adaptation (CAP-101)

Detection threshold: 3 occurrences of same pattern type within 14 days triggers suggestion

Overview

Knowledge Patterns extend the n-gram pattern detection (CAP-010) to capture conceptual learning signals, not just workflow sequences. A user repeatedly struggling with the same concept is as valuable a signal as a user repeatedly executing the same command sequence — but the original pattern detector couldn't see it.

Four pattern types are defined to cover the main conceptual learning signals: concept-usage (a concept appearing repeatedly across sessions), procedure-sequence (a domain-specific workflow repeated), error-correction (the same type of mistake recurring), and question-answer (the same question asked in different ways across sessions). Each type maps to a different kind of knowledge pack improvement — a new exercise, a captured workflow, a pitfall warning, or an explanatory section.

Knowledge patterns feed directly into Knowledge Adaptation (CAP-101), which uses the detected pattern types to personalise knowledge pack content for the individual learner. A learner who keeps triggering error-correction patterns for 15-bit arithmetic gets more scaffolding on that topic; a learner who never triggers that pattern skips the extra scaffolding and moves faster.

History

Pure sequence n-grams missed conceptual learning patterns — a user asking "what is copper list?" three different ways in three sessions was never detected as a pattern by the original CAP-010 pattern detector. The n-gram approach treated each question as a unique string and never noticed the semantic repetition.

Knowledge Pattern types were added specifically to capture these conceptual repetitions. The question-answer type was the hardest to implement — it requires semantic similarity detection rather than string matching, since "what is copper list?" and "how does copper list work?" are different strings but the same learning need. The solution was to embed knowledge pack concept labels (from the pack's concepts section) and match user queries against those embeddings, enabling fuzzy question-answer pattern detection across semantically similar queries.

Key Concepts

For Students & Teachers

Imagine a teacher who notices not just what a student does, but what they keep struggling with. If a student asks about the same concept three times in three different ways, a good teacher recognises that the concept isn't sticking and adjusts their teaching approach.

Knowledge Patterns give the system that same awareness. Instead of just tracking "what commands did you run?" (like the original pattern detector), it tracks "what concepts do you keep asking about?" and "what mistakes do you keep making?" These conceptual patterns are just as important for adaptive learning as workflow patterns are for skill generation.

Usage Examples

Example 1: Querying detected knowledge patterns

// Get all knowledge patterns for a learner in a domain
const patterns = await knowledgePatterns.getPatterns({
  learnerId: 'user-123',
  domain: 'apollo-agc',
  since: new Date('2026-01-01')
});

// Returns typed pattern records
patterns.forEach(p => {
  console.log(`Type: ${p.type}, Concept: ${p.concept}, Count: ${p.occurrences}`);
  // e.g. "Type: error-correction, Concept: 15-bit-arithmetic, Count: 4"
});

Example 2: Pattern triggering a knowledge pack improvement suggestion

Knowledge Pattern Alert
───────────────────────
Type:     question-answer
Concept:  copper-list-encoding
Domain:   amiga-heritage
Count:    3 occurrences (sessions: 2026-01-10, 2026-01-15, 2026-01-22)

Suggestion: Add "Copper List Encoding" as a new concept entry
            in the amiga-heritage knowledge pack.

Action: skill-creator suggest (to review this suggestion)