▶ Demo: the exam coach

study/
  exam-guide-v0.2.pdf     the official guide
  notes/                  your notes, ten modules

The coach drills you from your own material
and the guide's task statements. Nothing it
asks comes from anywhere else, so every
question is one the exam could ask.

# Task statements (Exam Guide v0.2)

## 5.2 Design effective escalation and
   ambiguity resolution patterns
Knowledge
- Escalation triggers: policy gaps, explicit
  request for a human, no meaningful progress
- Sentiment and self-reported confidence are
  unreliable proxies for case complexity
Skills
- Honor an explicit request for a human
  immediately, before investigating
- Ask for another identifier on multiple
  customer matches; never pick heuristically
Anti-patterns
- Confidence score as an escalation trigger

(29 more sections, same shape)

---
name: exam-coach
description: Drill the user for the Claude
  Certified Architect exam. Use when they say
  drill me, quiz me, or name a domain or task
  statement to practice.
---
1. Read study/task-statements.md and
   study/mistakes.md.
2. Pick 5 task statements: the ones named, or
   the domain named, weighted toward statements
   with misses in the log.
3. For each: write one scenario question in the
   guide's style (production symptom, four
   options, one correct, three plausible).
   Ask it. Wait for the answer. Do not reveal.
4. After the answer: say correct or wrong, give
   the reason in three sentences, and cite the
   task statement bullet it tests.
5. On a wrong answer, append one line to
   study/mistakes.md: date | TS | what was
   chosen | why it is wrong | the bullet.
6. End with a score out of 5 and the task
   statements to revisit.

---
name: weak-spots
description: Drill only the task statements the
  user has missed. Use when they say weak spots,
  drill my misses, or what am I getting wrong.
---
1. Read study/mistakes.md. Count misses per
   task statement.
2. Rank them. Take the top three (ties by most
   recent).
3. Run the exam-coach procedure on those three
   only, two questions each, with fresh
   scenarios (never reuse a logged question).
4. If a statement is answered correctly twice
   in a row across sessions, note it as
   "recovered" in the log. It leaves the top
   three; it does not leave the file.

---
name: readiness
description: Decide whether the user is ready
  to book the exam. Use when they say am I
  ready, readiness check, or paste domain scores.
---
1. Take the five domain percentages the user
   pastes from the study app and the mock.
2. Read study/mistakes.md; list statements
   missed twice or more without a "recovered"
   note.
3. Write study/readiness.md:
   - each domain, its score, and GO or NO GO
     against 85 percent
   - the overall verdict: GO only if every
     domain is GO and no statement is open
   - the checklist: booked, ID ready, room
     tested, sleep, two minutes a question,
     flag and move on
4. If NO GO: name the domains and statements
   to drill and stop there.

# Readiness  2026-09-09

| Domain | Score | Gate |
|---|---|---|
| 1 Agentic Architecture | 78% | NO GO |
| 2 Tool Design and MCP | 91% | GO |
| 3 Claude Code Config | 88% | GO |
| 4 Prompt Engineering | 86% | GO |
| 5 Context and Reliability | 90% | GO |

Open statements: 5.2 (2 misses), 1.5 (1 miss)

Verdict: NO GO. The average is 86.6 and the
exam does not grade on the average. Domain 1
is 27 percent of the paper. Drill 1.5, 1.6,
1.7, then rerun.

# Readiness  2026-09-11

| Domain | Score | Gate |
|---|---|---|
| 1 Agentic Architecture | 87% | GO |
| 2 Tool Design and MCP | 91% | GO |
| 3 Claude Code Config | 88% | GO |
| 4 Prompt Engineering | 89% | GO |
| 5 Context and Reliability | 92% | GO |

Open statements: none (5.2 recovered Sep 10,
1.5 recovered Sep 11)

Verdict: GO. Book it.
Checklist: [ ] booked  [ ] ID  [ ] room tested
[ ] sleep  [ ] 2 min a question  [ ] flag and
move on