pavankomateedi/synthetic-learner-red-team-harness

## 架构 ``` slh/ curriculum.py # research-grounded misconception catalog + problem bank personas.py # 8 archetypes; each differs in >=3 measurable dimensions learner.py # SyntheticLearner agent: seeded RNG + evolving memory model protocol.py # Move / Action / Turn / ItemState shared types tutor.py # TutorV1 (flawed baseline) + TutorV2 (improved policy) session.py # multi-turn runner: pre-assess -> teach -> post-assess detectors.py # avoidance / shallow-compliance / recovery detection evaluator.py # 7 PRD-8.1 dimensions, aggregate + per-persona comparator.py # before/after deltas, regression check, PRD-8.2 counter-metrics report.py # failure-mode + comparison Markdown renderers goldenset.py # the eval contract (passes via `slh check`) harness.py # orchestrates the recursive improvement loop cli.py # `slh` entry point ``` ## 可交付成果（PRD §12） | # | 可交付成果 | 文件 | |---|---|---| | 1 | 工作原型 | 本仓库 + `slh` CLI | | 2 | 合成学习者设计文档 | [docs/synthetic_learner_design.md](docs/synthetic_learner_design.md) | | 3 | 导师/课程文档 | [docs/tutor_and_curriculum.md](docs/tutor_and_curriculum.md) | | 4 | 评估方法 | [docs/evaluation_method.md](docs/evaluation_method.md) | | 5 | 基线与改进报告 | [docs/baseline_vs_improved.md](docs/baseline_vs_improved.md) | | 6 | 故障模式报告 | [docs/failure_report_baseline.md](docs/failure_report_baseline.md), [docs/failure_report_improved.md](docs/failure_report_improved.md) | | 7 | 递归改进描述 | [docs/recursive_improvement.md](docs/recursive_improvement.md) | | 8 | 决策日志 | [docs/decision_log.md](docs/decision_log.md) | | 9 | 研究笔记 | [docs/research_notes.md](docs/research_notes.md) | | 10 | 局限性备忘录 | [docs/limitations.md](docs/limitations.md) | | + | 金色集配套文件 | [docs/golden_set.md](docs/golden_set.md) |

标签：Apex, Markdown渲染, 人工智能, 决策日志, 命令行界面, 多轮交互, 局限性分析, 性能比较, 教学系统, 教育技术, 文档编写, 机器学习, 架构设计, 检测器, 模拟学习, 比较分析, 用户模式Hook绕过, 研究笔记, 记忆模型, 设计文档, 评估方法, 评估维度, 误解目录, 逆向工具, 递归改进, 递归改进循环, 问题库, 随机数生成, 黄金集