祝融说。

Exercise Overview

Purpose: Verify whether the trainee can independently complete the full "AI-assisted delivery" workflow -- from requirements analysis to blueprint design, from milestone breakdown to AI-assisted implementation, and from acceptance gating to retrospective summary.

Exercise format: Given a scenario modeled on realistic business conditions, the trainee independently completes every stage. AI coding tools are used throughout, but all decisions are made by the trainee. The instructor records the trainee's decision process and rationale.

Assessment dimensions: Score the five stages: requirements analysis, 20 points; blueprint design, 20; milestone breakdown and AI-assisted implementation, 30; acceptance and quality gating, 20; retrospective summary, 10. The total is 100 points. Comment on process completeness and decision soundness using evidence within the relevant stage; do not score them again separately.

Exercise Topics (Choose One of Four)

Topic A: Enterprise internal training system. Requirements: employees can publish courses and enroll in courses; instructors can view enrollment counts; administrators can manage users and course categories. Focus areas: role permissions, enrollment state machine, simple statistical reports.

Topic B: Pet boarding appointment platform. Requirements: users can select boarding services, view available time slots, pay a deposit online, and merchants confirm orders. Focus areas: appointment conflict detection, payment state machine, asynchronous communication between merchants and users.

Topic C: Factory equipment repair reporting system. Requirements: workers can submit equipment fault work orders; repair technicians can accept orders and fill in repair records; managers can view work order statistics. Focus areas: work order state machine, repair labor-hour statistics, role division.

Topic D: Community group-buying mini program. Requirements: group leaders can initiate group buys; residents can place orders to join groups; the platform can manage products and settlement. Focus areas: group-buy state transitions, inventory deduction, group leader reconciliation.

Completion Criteria (Grading Checklist)

The trainee must completely produce the following deliverables; the instructor scores each item against the checklist:

Stage 1: Requirements Analysis (20 points)

• □ List the core business event flow using Event Storming (in past tense);
• □ Derive commands, aggregates, and bounded contexts;
• □ Produce REQUIREMENTS.md (project overview, user roles, feature list, data entities, constraints);
• □ Record at least one "divergence record" (divergence, decision, reason, scope of impact, date).

Stage 2: Blueprint Design (20 points)

• □ Produce CONTEXT.md, fully containing seven parts: project overview, core glossary, tech stack, data model, API contracts, directory structure, milestone dependency tree;
• □ The glossary comes first and is precisely defined;
• □ The tech stack has clear recommendations + rationale + alternatives (proposing rather than asking);
• □ Milestones are "independently verifiable engineering nodes" (not vague feature points from a requirements perspective).

Stage 3: Milestone Breakdown and AI-Assisted Implementation (30 points)

• □ Each milestone is issued with an instruction containing the "four elements" (what to do / requirement details / technical constraints / acceptance criteria);
• □ Strict execution: proceed to the next milestone only after the current one is complete;
• □ During the coding phase, maintain "observation" rather than "micromanagement";
• □ Each milestone acceptance uses a structured acceptance report (functionality / code / architecture / security);
• □ Branch decisions are decisive: PASS commits, NEEDS_FIX fixes (at most twice), REBUILD rolls back and rebuilds;
• □ Experience at least one actual architecture drift and handle it correctly (identify -> escalate instruction -> rebuild).

Stage 4: Acceptance and Quality Gating (20 points)

• □ Each milestone has an acceptance record;
• □ Use high-frequency git commits (at least one commit per milestone);
• □ Core modules have unit tests and coverage meets the standard;
• □ Upon completion, CHANGELOG.md and CONTEXT.md are updated (documentation and code mirror each other).

Stage 5: Retrospective Summary (10 points)

• □ The retrospective report answers three questions: which decisions were done well / which decisions could be better / what would be done differently if redone once;
• □ Summarize one's 3 most frequently used skills and 1 skill most desired for improvement.

Exercise Report Template

# Final Comprehensive Exercise Report

## Project Name: ______

## I. Requirements Analysis
(Attach the Event Storming event flow, REQUIREMENTS.md, and divergence records.)

## II. Blueprint Design
(Attach the full CONTEXT.md.)

## III. Milestone Execution Records
| Milestone | Instruction highlights | Acceptance result | Commit |
|--------|---------|---------|------------|
| M1 | ...... | PASS | hash1 |
| M2 | ...... | NEEDS_FIX → PASS | hash2 |
| ...... | ...... | ...... | ...... |

## IV. Architecture Drift Handling Records (If Any)
- Drift signal: ______
- Higher-level instruction: ______
- Rebuild action: ______

## V. Retrospective Summary
1. Decisions made well: ______
2. Decisions that could be better: ______
3. What I would do differently: ______
4. My 3 most frequently used skills: ______
5. The 1 skill I most want to improve: ______

## VI. Instructor Comments by Stage
- Stage 1 · Requirements analysis: ___ / 20; evidence and comments: ______
- Stage 2 · Blueprint design: ___ / 20; evidence and comments: ______
- Stage 3 · Milestone breakdown and AI-assisted implementation: ___ / 30; evidence and comments: ______
- Stage 4 · Acceptance and quality gating: ___ / 20; evidence and comments: ______
- Stage 5 · Retrospective summary: ___ / 10; evidence and comments: ______
- Total: ___ / 100

Grading Scale

Internal completion certification in this exercise is only an assessment of learning in this training program. It is not an industry qualification certification.