Skip to content

FAQ Coverage Gaps

This report identifies concepts from the learning graph (475 total) that are not directly named in an FAQ question or answer. Many gap concepts are leaf-node terminology fully defined in the glossary — they don't necessarily need a dedicated FAQ entry. The gaps below are prioritized by their likely value as standalone FAQ questions.

Critical Gaps (High Priority)

These are high-centrality concepts (many dependencies in the learning graph or named explicitly in course outcomes) that deserve a dedicated FAQ entry:

  1. Tool Use / Function Calling
  2. Concept IDs: 92 (Claude Tool Use), 109 (Function Calling), 134 (Gemini Function Calling)
  3. Suggested question: "What is tool use and how does it affect token cost?"

  4. Category: Core Concepts

  5. Eval Suite / Golden Test Set

  6. Concept IDs: 466 (Eval Suite), 467 (Golden Test Set)
  7. Suggested question: "How do I design an eval suite for cost-quality benchmarking?"

  8. Category: Best Practice or Advanced

  9. Quality Regression Detection

  10. Concept ID: 453
  11. Suggested question: "How do I detect a quality regression in production?"

  12. Category: Best Practice

  13. Implicit vs Explicit Caching

  14. Concept IDs: 327, 328
  15. Suggested question: "What is the difference between implicit and explicit caching?"

  16. Category: Technical Details

  17. Reasoning Effort Setting / Verbosity Parameter

  18. Concept IDs: 391, 400
  19. Suggested question: "How do I control reasoning effort on the o-series and Claude extended thinking?"
  20. Category: Technical Details

Medium Priority Gaps

Concepts with moderate centrality that would round out the FAQ:

  1. CUPED Adjustment (Concept ID 284) — Advanced statistical technique for variance reduction
  2. Multi-Armed Bandit (Concept ID 283) — Sequential testing alternative to A/B
  3. Idempotency Key / Retry Policy (Concept IDs 433, 434) — Production-critical for batch jobs
  4. Backoff Strategy (Concept ID 435) — Retry economics
  5. Webhook Notification (Concept ID 432) — Batch result delivery
  6. Hybrid Retrieval / BM25 (Concept IDs 344, 345) — RAG patterns
  7. HyDE / Query Rewriting (Concept IDs 347, 348) — Advanced RAG
  8. Subagent Pattern (Concept ID 165) — Harness architecture
  9. Persistent Memory File (Concept ID 167) — Agent memory patterns
  10. Token Spike Alert / Cost Threshold Alert (Concept IDs 235, 236) — Alerting design
  11. Cardinality Concern (Concept ID 237) — Dashboard footgun
  12. Pareto Analysis / Top-N Cost Drivers (Concept IDs 243, 244) — Log analysis methodology
  13. 5-Hour Limit / Weekly Limit (Concept IDs 471, 472) — Vendor-imposed rate limits
  14. Parallel Token Penalty (Concept ID 475) — Why parallel execution can cost more
  15. Continuous Cost Monitoring / Token Efficiency Roadmap (Concept IDs 469, 470) — Programmatic monitoring

Low Priority Gaps

Leaf-node terminology already well covered in the glossary; doesn't need a dedicated FAQ entry unless reader feedback indicates demand:

  • Special Tokens, EOS Token, BOS Token, Padding Token (39-42)
  • Whitespace Handling, Unicode Normalization (37, 38)
  • Token Boundary, Pre-Tokenization (43, 44)
  • Logit Bias, Seed Parameter (119, 120)
  • Volatile Suffix, Cache Boundary (320, 321)
  • Lost-In-The-Middle, Context Reordering (367, 368)
  • Pilot Rollout, Canary Deployment (457, 458)
  • Vendor Data Retention, Opt-Out Of Training (444, 445)
  • Tokenized Identifier (448)

Recommendations

  1. Address all 5 critical gaps in the next FAQ revision — these are concepts named in course learning outcomes that readers will search for by name.
  2. Add 5–8 medium-priority entries to push coverage above 50% directly named.
  3. Low-priority gaps can stay glossary-only; the glossary is the right place for narrow terminology with a one-sentence definition.

Coverage Summary

Priority Gap Count Action
Critical 5 Add to FAQ in next revision
Medium 15 Consider for FAQ; OK in glossary
Low ~270 Glossary entries sufficient
Total gap ~290 / 475 Reflects glossary-as-canonical strategy

The FAQ is sized appropriately for a professional-development textbook (91 questions). Pushing it past 130 questions without strong reader demand risks dilution; instead, expand the glossary and rely on search for the long tail.