89 / 90

第 89 天:糾正、部分失敗回饋與 Agent 輔助工程檢查

把第 88 天的未達項轉化為已驗證的糾正,並在 Capstone Agent 專案上完成一個有邊界的委派任務——輔助證據嚴格分離。

Agent 與智能體工作流 480 分鐘有效研習(8 小時)
01 120 分鐘

部分失敗分類法與糾正協定

Agent 輔助工程檢查

前置要求

第 88 天評分表與未達標台帳。

學習目標

把每個低於 2 分的量規項分類為 concept_gap、slip、ambiguity 或 evidence_gap,並應用五步糾正協定:複述、定位、最小復現、修復、複測。

操作指引

1. 四種未達類型及其特徵:
- concept_gap:完全無法產出方法(如列不出連鎖律)。走對映的恢復日——重新學習後挑戰*新的*變體任務。
- slip:方法正確但算術/細節出錯(如符號錯、差一)。直接複測同任務,無需完整恢復日。
- ambiguity:任務措辭不清。記錄不清之處,複測前先澄清。
- evidence_gap:答案正確但未執行驗證命令。補出命令輸出即可,無需重學。
2. 對每列台帳應用糾正協定:(a) 用自己的話複述任務;(b) 精確定位失敗點;(c) 構建展示該失敗的最小復現;(d) 只修這一點;(e) 複測並記錄驗證輸出。
3. 回饋品質規則:每條自我批註必須引用失敗的量規準則,而非只寫「錯了」。例:「PY-1 未達準則 tie-order:most_common 回傳插入序;修復=按(-count, word)排序」。
4. 逐列判斷:走完整恢復日(concept_gap)還是 20 分鐘定向複習(slip、evidence_gap)。

自主實作與記錄

產出 corrections_plan.json:為第 88 天每列台帳標註 miss_type、未達量規準則、已做糾正步驟、複測命令或變體任務。

獨立掌握標準

每個失敗項都誠實分類並帶 miss_type、未達準則、糾正步驟與複測計畫。

權威來源資源與權利證明

linking Journal publisher copyright (all rights reserved) - lawful external linking only

Improving Students' Learning With Effective Learning Techniques: Promising Directions From Cognitive and Educational Psychology

作者 / 機構: John Dunlosky, Katherine A. Rawson, Elizabeth J. Marsh, Mitchell J. Nathan, Daniel T. Willingham | 原始出處連結 ↗

對應章節 / 時間戳: Psychological Science in the Public Interest 14(1): 4-58; practice testing and distributed practice sections

驗證版本與日期: Psychological Science in the Public Interest, 2013 (Verified: 2026-10-30)

入選理由:

Detailed comparative review rating practice testing and distributed practice as high-utility techniques while rating highlighting and rereading low-utility; anchors the retrieval, correction, and spaced-review design of Days 88-90.

來源前置條件:

None beyond general academic reading.

安全回退方案(離線彈性):

Curriculum restates the utility ratings and application conditions in original prose; every task remains runnable offline.

對比候選資源與選型分析 (2)
Pashler et al. Learning Styles: Concepts and Evidence ↗ compared_alternative [Publisher copyright (originally Wiley; now SAGE Publications)]

Important negative result on learning-styles matching but narrower scope; Dunlosky et al. rates a broader set of techniques directly usable for this curriculum.

Brown, Roediger & McDaniel: Make It Stick: The Science of Successful Learning ↗ compared_alternative [Publisher copyright (Harvard University Press)]

Strong trade-book synthesis of the same evidence base; the peer-reviewed review is preferred for citation precision and technique-by-technique utility ratings.

linking PSF License Version 2

Python Standard Library: Built-in Exceptions

作者 / 機構: Python Software Foundation | 原始出處連結 ↗

對應章節 / 時間戳: Concrete exceptions and Exception hierarchy

驗證版本與日期: Python 3.12 (Verified: 2026-10-30)

入選理由:

Normative exception hierarchy used to justify the narrowest correct except clause and FileNotFoundError attributes such as filename and errno.

來源前置條件:

Basic Python syntax from Study Days 01-09

安全回退方案(離線彈性):

The drill reproduces FileNotFoundError and ValueError locally, and help(ValueError) documents the hierarchy offline.

對比候選資源與選型分析 (2)
W3Schools Python Exceptions ↗ compared_rejected_for_days10_12 [Proprietary]

Beginner-friendly but non-normative; omits the OSError attributes and hierarchy precision the diagnosis drills require.

Broader class-oriented chapter; the dedicated exceptions reference is the tighter fit for reading and diagnosing concrete error types during debugging checks.

02 120 分鐘

定向糾正與複測(無輔助)

獨立掌握檢查(無 AI 輔助)

前置要求

第 89 天活動 1 的糾正計畫;輔助條件:獨立。

學習目標

在相同無輔助條件下重做第 88 天全部失敗任務,用可執行命令驗證糾正,並誠實複評——前後對比,而非覆蓋。

操作指引

1. 每個 concept_gap 列:先完成對映的恢復活動,然後挑戰*變體*任務——不是原題,讓複測度量遷移而非背答案。
2. 每個 slip 列:重答原任務並把兩次作答做 diff。
3. 必須執行並記錄的驗證命令:梯度答案→float64 下 h=1e-5 的雙側有限差分 (f(x+h)−f(x−h))/(2h);git 序列→在一次性複製上執行並展示 git log --oneline;Python 函式→python -m unittest -v;參數計數→重算並與 projects/pretrain/artifacts/run_artifact.json 做 diff。
4. 記錄 corrections.jsonl——每個複測項一列:{task_id, before, after, verify_command, verified: true|false}。
5. 仍低於 2 分的保持低分——保持低分——記錄具名恢復路徑,絕不靜默升級。

自主實作與記錄

輸出覆蓋第 88 天全部台帳列的 corrections.jsonl,每列含前後得分與實際捕獲的驗證輸出。

獨立掌握標準

每個複測項要麼是命令驗證過的糾正,要麼是帶具名恢復路徑的明確未達——不得靜默升級。

權威來源資源與權利證明

linking MIT

CS231n: Backpropagation and Intuitions

作者 / 機構: Andrej Karpathy / Stanford University | 原始出處連結 ↗

對應章節 / 時間戳: Patterns in backward flow

驗證版本與日期: Course notes (Verified: 2026-10-30)

入選理由:

Modular local-gradient interpretation (gate operations: add, multiply, max) and gradient flow intuition.

來源前置條件:

Multivariable calculus and basic Python.

安全回退方案(離線彈性):

Local gate definitions and circuit flow diagrams in the lesson provide full coverage.

對比候選資源與選型分析 (2)

Raw circuit view; CS231n notes are more structured with explicit gate derivations.

Fast.ai Practical Deep Learning for Coders ↗ compared_alternative [GPL-3.0]

Top-down approach; CS231n bottom-up modular view matches our first-principles curriculum.

linking CC BY-NC-SA 3.0 noncommercial share-alike

Pro Git, Second Edition

作者 / 機構: Scott Chacon and Ben Straub | 原始出處連結 ↗

對應章節 / 時間戳: Chapter 3 Git Branching and Section 7.7 Reset Demystified

驗證版本與日期: Second edition (Apress, 2014) (Verified: 2026-10-30)

入選理由:

Freely licensed reference for branching, merging, conflict resolution, and safe recovery vocabulary used on Day 14.

來源前置條件:

Basic git status/log/diff vocabulary from earlier study days

安全回退方案(離線彈性):

git help <command> and the Git manual pages ship with Git and cover the same commands offline; every required workflow is stated locally in the lesson.

對比候選資源與選型分析 (2)
GitHub Skills interactive courses ↗ compared_rejected_for_day14 [MIT (varies by course)]

Interactive but requires a GitHub account and an external service; disposable local repositories keep the exercise self-contained.

Atlassian Git Tutorials ↗ compared_rejected_for_day14 [All Rights Reserved]

Vendor-owned prose with no permissive license; Pro Git covers the same ground under CC BY-NC-SA.

03 120 分鐘

Agent 輔助工程檢查:在 Capstone Agent 上的有邊界委派任務

Agent 輔助工程檢查

前置要求

第 89 天活動 2;輔助條件:agent_assisted。本項是 Agent 輔助工程檢查——其證據絕不轉化為獨立掌握。

學習目標

在 Capstone 專案上證明合格的 Agent 指揮能力:撰寫含驗收標準與權限範圍的有邊界規格,在一次性分支上套用變更,審查 diff 是否越界,執行確定性評測,並記錄誠實結果。

操作指引

1. 在 projects/agent_learning 上選一個有邊界任務,如給 scorer CLI 增加 --json 輸出旗標,或給 summarize-notes fixture 增加一條筆記。
2. 委派前先寫規格:目標、Agent 可觸碰的檔案(如 cli.py 與一個測試檔案)、驗收標準(python -m unittest tests.projects.test_agent_learning 通過;新旗標輸出合法 JSON)、停止條件。
3. 把 Agent 的修補套用到一次性分支(自己的複製內 git checkout -b delegated-task 或暫存副本);不得用於受保護工作樹。
4. diff 審查清單:無聲明範圍外檔案;未新增網路呼叫;無金鑰憑據;測試真正覆蓋新行為;錯誤路徑保留。
5. 執行驗證:專案自身測試/評測命令+你的驗收檢查。記錄:規格文本、變更檔案、diff 審查要點、驗證命令與輸出、Agent 迭代次數、你的判定(accepted / accepted-with-fixes / rejected)。

自主實作與記錄

輸出 assisted_check.json:{spec, files_changed[], diff_review_notes, verify_command, verify_output_summary, agent_iterations, verdict}。

獨立掌握標準

端到端完成一個委派任務,含已審查 diff、已驗證測試執行、誠實迭代計數與有據判定——標註為輔助證據。

權威來源資源與權利證明

linking Open Access / CC BY 4.0

ReAct: Synergizing Reasoning and Acting in Language Models

作者 / 機構: Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, Yuan Cao | 原始出處連結 ↗

對應章節 / 時間戳: Section 3 & 4 (ReAct Formulation and Synergy)

驗證版本與日期: ICLR 2023 / arXiv:2210.03629v2 (Verified: 2026-10-30)

入選理由:

Foundational academic paper introducing the interleaved Thought-Action-Observation control loop in LLM agents, establishing formal interleaving of task reasoning and external tool execution.

來源前置條件:

Prompt engineering, token generation basics, and conversational interaction loops.

安全回退方案(離線彈性):

Complete algorithmic step traces, state loop diagrams, and control loop pseudocode are self-contained in curriculum materials.

對比候選資源與選型分析 (2)

Pure reasoning without environmental actions or external tool grounding; ReAct extends reasoning into an interactive environment action-observation loop.

Early neuro-symbolic modular router concept; ReAct offers simpler, prompting-native action-observation loop dynamics suitable for standard API models.

linking Open Access / CC BY 4.0

SWE-bench: Can Language Models Resolve Real-World GitHub Issues?

作者 / 機構: Carlos E. Jimenez, John Yang, Alexander Wettig, Shunyu Yao, Kexin Pei, Ofir Press, Karthik R. Narasimhan | 原始出處連結 ↗

對應章節 / 時間戳: Section 2.1 Benchmark Construction and Section 2.2 Task Formulation (FAIL_TO_PASS tests)

驗證版本與日期: ICLR 2024 / arXiv:2310.06770v3 (Verified: 2026-10-30)

入選理由:

Defines the standard pattern this block teaches: a bounded issue statement paired with executable fail-to-pass tests as verifiable acceptance criteria for agent-produced changes.

來源前置條件:

Unit testing, issue descriptions, and repository evaluation vocabulary.

安全回退方案(離線彈性):

The acceptance-criteria pattern is exercised locally through the Ticket 22 fixture scorer and the Ticket 24 regression exercise without external datasets.

對比候選資源與選型分析 (2)
HumanEval: Evaluating Large Language Models Trained on Code ↗ compared_alternative [arXiv non-exclusive license / Open Access]

Function-level synthesis benchmark; SWE-bench models the repository-scoped issue-to-diff workflow that this delivery block rehearses.

Multi-SWE-bench: A Multilingual Benchmark for Issue Resolving ↗ compared_alternative [Open Access / CC BY 4.0]

Extends issue resolution to multiple programming languages; the original Python-centric SWE-bench provides a cleaner pedagogical anchor.

04 120 分鐘

複習:補救台帳收尾與誠實分數更新

獨立掌握檢查(無 AI 輔助)

前置要求

第 89 天活動 1-3 完成。

學習目標

閉環:用糾正後的前後結果更新掌握台帳,讓仍未達項保持可見未達,並把輔助證據記入永不併入獨立分數的單獨欄。

操作指引

1. 合併評分表:每個域按第 88 天表與今日複測寫出前後小計。
2. 仍 <2 分的項連同具名恢復路徑結轉到第 90 天後續學習計畫——保持可見未達。
3. 活動 3 的判定記入標註 Agent 輔助工程檢查的 assisted_evidence 欄;它證明委派能力,絕非無輔助知識。
4. 寫最終證據說明:哪些工件經命令驗證(獨立),哪些是自我報告或輔助;列出產生驗證的準確命令。
5. 輸出 mastery_ledger.json:各域前後、帶恢復路徑的剩餘缺口、independent_evidence 清單、assisted_evidence 清單、誠實範圍聲明。

自主實作與記錄

產出填寫完整的 mastery_ledger.json;輔助結果只能出現在 assisted_evidence 下。

獨立掌握標準

最終台帳誠實分離獨立與輔助證據,未達標準保持可見未達,範圍聲明屬實。

權威來源資源與權利證明

linking Journal publisher copyright (all rights reserved) - lawful external linking only

Test-Enhanced Learning: Taking Memory Tests Improves Long-Term Retention

作者 / 機構: Henry L. Roediger III, Jeffrey D. Karpicke | 原始出處連結 ↗

對應章節 / 時間戳: Psychological Science 17(3): 249-255, Experiments 1-2

驗證版本與日期: Psychological Science, 2006 (Verified: 2026-10-30)

入選理由:

Foundational experimental evidence that retrieval practice (testing) outperforms repeated study for long-term retention; grounds the Day 88 AI-free retrieval checks in peer-reviewed learning science rather than folk advice.

來源前置條件:

Basic understanding of study sessions, retention intervals, and experimental design vocabulary.

安全回退方案(離線彈性):

The curriculum restates the experimental design and effect sizes in original prose; every sealed task runs without the paper.

對比候選資源與選型分析 (2)

Strong meta-analytic coverage of spacing effects but addresses scheduling rather than the retrieval effect itself; retained as support for the Day 90 spaced-review schedule.

Equally foundational follow-up demonstrating repeated retrieval vs repeated study; the 2006 paper is cited as the standard first controlled demonstration.

linking CC BY 4.0

The Turing Way: A Handbook for Reproducible, Ethical and Collaborative Data Science

作者 / 機構: The Turing Way Community | 原始出處連結 ↗

對應章節 / 時間戳: Reproducible Research chapter: version control, testing, environment capture

驗證版本與日期: Zenodo DOI 10.5281/zenodo.3233853, continuously updated (Verified: 2026-10-30)

入選理由:

Community-maintained CC-BY handbook codifying reproducibility checklists (versioned code, environment capture, rerunnable commands); directly defines the Day 90 reproduction-verification and provenance-audit workflow.

來源前置條件:

Basic version control and command-line familiarity.

安全回退方案(離線彈性):

The portfolio checklist and verification log schema are authored locally; reproduction commands run without the handbook.

對比候選資源與選型分析 (2)

Concise practices paper; The Turing Way's maintained, checklist-driven coverage of environments and reproducibility is broader and kept current.

Foundational rules list; The Turing Way expands the same rules into actionable checklists used verbatim for the Day 90 verification run.