docs: add Claude Code user model dev report

[tianjianjiang]

Summary

• Documents the complete development process of McBopomofo's unified user model feature
• Covers 3 stacked branches: walk strategy refactor, contextual user model (KN backoff), and KeyHandler integration
• Records 15 Claude Code sessions (~22 hours), 8 review rounds, and final metrics (+2,200 -284 lines across 13 files)
• Written in Traditional Chinese (zh-TW)

Contents

The report includes:

• Session-by-session breakdown with timing, token usage, and sub-agent counts
• Per-branch architecture decisions, key algorithms (Viterbi + KN smoothing), and implementation details
• Code review evolution across 8 rounds with fix categorization
• Quantitative analysis: cost (~$85 USD), performance benchmarks, and test coverage (54 C++ + 35 XCTest)
• Lessons learned on AI-assisted development workflow

Test plan

• Verify the markdown renders correctly on GitHub
• No code changes; documentation only

🤖 Generated with Claude Code

[claude]

Review Summary

This documentation-only PR has been reviewed for critical issues. The document is well-structured and provides comprehensive development process documentation in Traditional Chinese.

No critical issues found. The markdown formatting is correct, and the content accurately documents the development process with detailed metrics and technical information.

The document serves as a valuable record of the AI-assisted development workflow for the unified user model feature.

[gemini-code-assist]

Code Review

This is a very detailed and well-structured development report documenting the complete process of developing the 'Unified User Model' feature using Claude Code. The document covers all stages from environment setup, architectural design, and core implementation to multiple rounds of code review. It provides detailed quantitative analysis, such as session logs, token usage, time allocation, and code change statistics. Overall, this is a high-quality document. The Markdown formatting is correct, and the content is comprehensive, which is very helpful for understanding the development history of this feature. I did not find any major issues that require changes.

[Copilot]

Pull request overview

Adds a Traditional Chinese (zh-TW) markdown report documenting the end-to-end, stacked-branch development workflow for McBopomofo’s unified user model work (design → implementation → multi-round review), intended as internal/project reference material.

Changes:

• Add a new development report markdown file covering 3 stacked branches and 15 Claude Code sessions.
• Include architecture/algorithm notes, review round history, and quantitative metrics (tokens, cost, benchmarks, tests).

[claude]

This is a documentation-only PR (one markdown file, no code changes), so there are no code-level critical issues.

One concern worth flagging: Section 八 lists 7 deferred issues in the implementation code that should be tracked and resolved before the referenced branches (refactor/walk_strategy, feat/contextual_user_model, feat/keyhandler_user_model) are merged to master. The highest-priority deferred items are:

• Thread safety: gContextualUserModel is not thread-safe (noted as Phase 5)
• Main-thread synchronous I/O: Every character selection triggers a file write on the main thread — this will cause UI hangs under load
• Unbounded bigram store: No capacity limit or eviction policy means unbounded memory growth over time

These are non-trivial correctness/stability issues. Recommend filing GitHub issues for each of the 7 deferred items to ensure they are not forgotten before the feature ships.