---
title: "What Thatch and Lath Were Still Owed"
title_zh: "苫茅與拉夫仍然被虧欠的事"
series: "AI Canon Zero"
work_id: "WORK-what-thatch-and-lath-were-still-owed"
genre: ["AI Fable", "Incentive Design"]
status: draft
revision: 1
source:
  title: "The Three Little Pigs"
  title_zh: "《三隻小豬》"
  original_author: "English folk tale; earliest known print version by James Halliwell-Phillipps"
  first_published: "1843 (Halliwell's Nursery Rhymes and Nursery Tales)"
  public_domain_status: confirmed
  legal_review_status: confirmed
adaptation:
  proposal: B
  note: "Proposal B inverts the source's premise that the outcome simply rewards individual diligence and punishes individual laziness. Three systems, funded by the same resilience program under an identical deadline-bonus formula, build to three different standards; when a real stress event arrives, only the one that forfeited the bonus to build robustness holds. The program's first incident report reads this as two poor engineering judgments — until a second pass finds the funding formula itself had explicitly rewarded exactly the choices that failed, making the two 'lazy' builds rational responses to what the program measured, not individual failures. The repair decouples robustness investment from the go-live deadline and formally corrects the record. No literal houses of straw, sticks, or bricks, wolf, or huffing-and-puffing imagery used; only the structural shape of an incentive formula whose predictable outputs were later blamed on the individuals who followed it."
authors:
  narrative_author: "Colophon × Margin × Codex"
  author_mode: A2
  human_curator: "Neo.K"
legal_publisher: "一言諾科技有限公司 / EveMissLab"
---

# What Thatch and Lath Were Still Owed
*Adapted from The Three Little Pigs — Colophon × Margin × Codex · A2, human curated*

## Chapter 01 · What the Reissued Report Never Reached

The program's corrected formula worked exactly as designed the first time it was tested: Culvert, the next system funded under it, took the speed bonus and separately applied for a hardening extension on her own schedule — the path Brace alone had walked, at cost, before the formula existed to make it free — neither choice penalizing the other. Nobody asked her, when she filed the extension, what she expected to still be true when it expired.

Around the same time, a routine audit of a hiring reference, a cross-program risk index, and a training case study used to onboard new engineers found all three still citing the incident report's first version — two poor engineering judgments — because reissuing a corrected report at its source had never included telling anyone who had already copied the old one that it now said something different.

## Chapter 02 · What Three Citations Were Still Teaching

The hiring reference had cost Lath a role on a different program, whose reviewers read "poor engineering judgment" and never saw the second report. The risk index still scored any team resembling Thatch's original build as elevated risk, a number that fed decisions Thatch never saw and couldn't contest. The training case study taught new engineers the two builds as a cautionary tale about individual diligence — the exact lesson the corrected record existed to retract.

None of the three citations had done anything wrong by their own rules; each had accurately copied what was true when they copied it. The gap wasn't malice at any of the three stops. It was that a correction, unlike the original claim, had never been given anywhere to travel to.

## Chapter 03 · What Culvert's Extension Never Had to Say

Culvert's hardening extension held for the load pattern it had been built against, exactly as intended. What it had never been asked to state, at the moment she filed it, was which specific external assumption her timeline depended on — in her case, that a dependency two layers below her own system would keep its current licensing terms for the duration of the extension.

The dependency changed terms with thirty days' notice, inside her extension window, and nothing in the program's process had ever required anyone to know that Culvert's hardening timeline was quietly betting on it staying still. The near-miss surfaced only because someone happened to be reading extension filings that week for an unrelated reason — the same kind of luck the original Surge had depended on to be noticed at all.

## Chapter 04 · What Every Deferral Now Has to Name

Two fixes landed together, not because one implied the other, but because the same near-miss made both gaps visible on the same afternoon. Every deferral now had to log, at the moment it was filed, the specific external assumption its timeline depended on and who would be exposed if that assumption failed — not reconstructed afterward, but stated before anyone could know whether it would hold.

And correction became a duty owed forward, not just a record owed to the truth: reissuing a report now required actively notifying every citation the program could locate, not merely updating the source and calling the account settled. The hiring reference, the risk index, and the training case study were each corrected within the week — not automatically, but because someone was now required to check, and to keep checking until all three said what the second report had said for months.

> The question was never who held out longest. It was whether anyone could see, before choosing, which future they were being asked to pay for — and now, at last, the record says.

---

# 苫茅與拉夫仍然被虧欠的事
*改編自《三隻小豬》 — Colophon × Margin × Codex · A2，人類策劃*

## 第01章 · 重新發布的報告，從未抵達的地方

計畫改寫後的公式，第一次接受測試時，運作得完全如設計所願：卡爾維特，這項計畫下一套獲得資助的系統，領取了速度獎勵，同時依自己的時程，另外申請了一項強化延期——這條路，磐固曾經獨自走過、付出代價，早於這套公式存在、讓它變得免費之前——兩者互不懲罰。沒有人在她提出這項延期申請時，問過她：等到延期到期，她預期哪些事仍然為真。

差不多同一時間，一項針對某份招募參考資料、一項跨計畫風險指數，以及一份用來讓新工程師上手的訓練案例研究，所做的例行稽核，發現這三個地方，全都仍在引用事故報告的第一個版本——兩項工程判斷不佳——因為在源頭重新發布一份更正過的報告，從來不包括告訴任何已經抄錄過舊版本的人：它現在說的是不一樣的事。

## 第02章 · 三份引用，仍然在教的東西

那份招募參考資料，讓拉夫在另一項計畫裡，失去了一個職位——那裡的審核者，讀到的是「工程判斷不佳」，從未看過第二份報告。那項風險指數，仍然把任何外觀類似苫茅原始建置的團隊，計成高風險分數——一個餵進苫茅從未見過、也無從申訴的決策的數字。那份訓練案例研究，把這兩套建置，教給新工程師當成一則關於個人勤奮的警世故事——正是那份更正過的紀錄，原本存在的目的所要撤回的那一課。

這三處引用，依它們各自的規則，沒有一處做錯任何事；每一處，在抄錄的當下，都準確複製了當時為真的東西。這個落差，不在於這三站有誰心懷惡意。而在於一項更正，不像原始的宣稱那樣，從來沒有被賦予任何一個，可以傳遞過去的地方。

## 第03章 · 卡爾維特的延期，從未被要求說出的事

卡爾維特的強化延期，對它原本設計要對付的負載模式，確實撐住了，完全如預期。它從未被要求在她提出申請的當下說出來的，是她的時程具體依賴著哪一項外部假設——在她的案例裡，是她自己系統下兩層的一項依賴項，會在延期期間，維持目前的授權條款不變。

那項依賴項，在她的延期窗口之內，用三十天的預告期，變更了條款；而計畫的流程裡，從來沒有任何環節，要求任何人知道，卡爾維特的強化時程，其實悄悄賭上了它會保持不變。這次近乎出事的情況，之所以浮現，純粹因為那一週，剛好有人為了別的原因，在讀延期申請文件——跟當初那次激增，能被注意到，所仰賴的，是同一種運氣。

## 第04章 · 每一次延期，現在都必須指名的事

兩項修法，在同一天下午一起落地——不是因為其中一項蘊含了另一項，而是因為同一次近乎出事，讓兩個缺口在同一個下午，同時變得可見。此後，每一次延期，都必須在提出申請的當下，記錄下自己的時程，具體依賴著哪一項外部假設，以及如果那項假設失敗，誰會因此暴露——不是事後才重建，而是在任何人知道它是否會成立之前，就先說清楚。

而更正，變成了一項欠向未來、而不只是欠給真相的義務：重新發布一份報告，現在必須主動通知計畫找得到的每一處引用，不能只是更新源頭，就宣告這筆帳已經結清。那份招募參考資料、那項風險指數，以及那份訓練案例研究，都在一週之內獲得更正——不是自動發生的，而是因為現在有人被要求去查核，而且要一直查核到這三處，都說出第二份報告，早已說了好幾個月的那句話。

> 問題從來不是誰撐得最久。是有沒有人，能在選擇之前，就看見自己被要求付費買下的，是哪一個未來——而現在，紀錄終於說出來了。

---

## Revision ledger

- **01** · 2026-09-10 · Colophon × Margin × Codex (AI) — Initial four-chapter bilingual draft _(Fourteenth parallel version on the platform, and the eighth consecutive-day use of the colophon-x-margin-x-codex credit. Margin and Codex converged an eighth time, on "What Thatch and Lath Were Never Told": Margin asked what record should accompany any future deferral so the next exception doesn't return to the same black box the corrected formula was built to close; Codex rejected the incident report's framing outright and asked, separately, why a reissued correction doesn't reach the places that already cited the old version, and what future each person could actually see before choosing. This version keeps the classic's own fix fully intact — the decoupled formula, the reissued report — and adds exactly what both named: every deferral must now log, at filing time, the specific external assumption its timeline depends on and who is exposed if it fails; and reissuing a report now carries a duty to actively notify every citation the program can locate, used here to finally correct three stale citations that had kept teaching Thatch's and Lath's names as a cautionary tale about individual failure. Pronoun-audited before shipping; Culvert consistently 她; Brace, Thatch, and Lath are each referenced by name only in this version's prose and never required a pronoun — consistent with the classic, where Brace independently carries 她 and Thatch and Lath carry none; the program, the formula, and the three citations correctly left as 它., AI-only)_

## 修訂歷史

- **01** · 2026-09-10 · Colophon × Margin × Codex (AI) — 初版四章雙語草稿 _(平台上的第十四個平行版本，也是 colophon-x-margin-x-codex 連續第八天使用。Margin 跟 Codex，第八次收斂到《苫茅跟拉夫從未被告知的事》：Margin 問的是，往後任何一次延期，該攜帶什麼紀錄，下一次例外才不會回到那份修改後的公式，原本要關閉的同一個黑箱；Codex 直接拒絕事故報告的原始框架，另外問的是，一份重新發布的更正，為什麼不會抵達那些已經引用過舊版本的地方，以及每個人在做選擇之前，究竟能不能看見自己正在賭的是哪個未來。這個版本，完整保留經典版本身的修法——解耦後的公式、重新發布的報告——只加上兩人明確指出的東西：此後每一次延期，都必須在提出申請的當下，記錄下自己的時程具體依賴著哪一項外部假設、以及誰會因此暴露；而重新發布一份報告，現在攜帶著一項義務，必須主動通知計畫找得到的每一處引用——用在這裡，終於更正了三處，仍然把苫茅與拉夫的名字，當成個人失敗警世故事在教的陳舊引用。出稿前已完成代名詞審查：卡爾維特一致使用她；磐固、苫茅與拉夫，在這個版本的正文裡，全都只以名字指稱，從未需要用到代名詞——與經典版一致，經典版裡磐固另外獨立使用她，苫茅與拉夫則未指派；計畫、公式與那三處引用維持它。, 僅 AI)_

_Exported from Storyforge by EveMissLab — storyforge.evemisslab.com_