Bank stats — CCA-F
Question inventory, coverage, and quality signals. Aggregate only — individual users' scores stay private.
Inventory
885
questions in the bank
| Verified | 870 |
| Awaiting review | 15 |
| Rejected (excluded from play) | 3 |
| Difficulty mix | 16 easy · 374 medium · 495 hard |
Usage
| Served at least once | 208 of 885 (24%) |
| Never seen by anyone | 677 |
| Study sessions | 53 |
| Answers recorded | 309 |
| Users | 13 |
| Open question flags | 4 |
By domain (bank share vs exam weight)
| Domain | Exam weight | Bank share | Verified | Unverified | Served | Miss rate |
|---|---|---|---|---|---|---|
| Agentic Architecture | 27% | 26.2% (232) | 230 | 2 | 57/232 | 16% of 77 |
| Claude Code Configuration | 20% | 20.3% (180) | 177 | 3 | 42/180 | 18% of 62 |
| Prompt Engineering | 20% | 20.8% (184) | 178 | 6 | 43/184 | 26% of 65 |
| Tool Design & MCP | 18% | 17.9% (158) | 157 | 1 | 29/158 | 20% of 45 |
| Context Management | 15% | 14.8% (131) | 128 | 3 | 37/131 | 19% of 47 |
⚠ = the bank is thinner than the exam weight suggests — generate or import more items there so mocks sample it realistically.
By author (who wrote the questions, and how they play)
| Author / model | Source | Questions | Answers seen | Miss rate |
|---|---|---|---|---|
chatgpt |
imported bank | 411 | 76 | 9% |
claude-fable-5 |
imported bank | 191 | 144 | 13% |
gemini |
imported bank | 96 | 11 | 36% |
olivier-alter-practice-exam |
imported bank | 77 | 15 | 33% |
expert-exam-author |
imported bank | 53 | 12 | 17% |
claude-platform-mechanics-docs |
imported bank | 24 | 8 | 38% |
mc-marcocheng-gist-cherrypick |
imported bank | 17 | 11 | 55% |
Qwen3.6-35B-A3B-GGUF |
generated in-app | 16 | 19 | 63% |
A source whose miss rate runs far above the rest may be writing harder — or lower-quality — questions. Cross-check with flags before judging.
Most-missed questions (min 2 answers)
| # | Question | Domain | Missed |
|---|---|---|---|
| 260 | A medical-coding assistant must abstain when documentation does not support a specific cod… | Prompt Engineering | 4/4 (100%) |
| 80 | ... | Claude Code Configuration | 2/2 (100%) |
| 47 | A high-volume LLM feature has p95 latency and cost problems. Profiling shows: 30k input to… | Context Management | 6/7 (86%) |
| 2 | A production data ingestion service parses unstructured customer support transcripts to ex… | Prompt Engineering | 3/4 (75%) |
| 33 | Reviewing a repo's Claude Code setup, you find these four items in CLAUDE.md. Which one ac… | Claude Code Configuration | 2/3 (67%) |
| 666 | Production data shows that in 12% of cases, your agent skips `get_customer` entirely and c… | Agentic Architecture | 2/3 (67%) |
| 753 | Production metrics show that when resolving complex billing disputes or multi-order return… | Prompt Engineering | 2/3 (67%) |
| 759 | After calling `get_customer` and `lookup_order`, the agent has all available system data b… | Context Management | 2/3 (67%) |
| 799 | A system provides three tools: `get_user_name`, `get_user_email`, and `get_user_address`. … | Tool Design & MCP | 2/3 (67%) |
| 3 | A travel-tech company uses an LLM to assist agents in booking complex multi-leg itinerarie… | Prompt Engineering | 1/2 (50%) |