lilmgenius/paperthin

mandela

Audit any eval, metric, experiment, or benchmark for leakage: does external ground-truth enter independently, or are the model, scorer, and designer just confirming a result no outside truth ever produced?

查看源码

使用方法

  1. 复制安装命令,在自己的智能体环境安装前先检查源码包。
  2. 从原始文档核对所需工具、凭据和支持的宿主。
  3. 说明输入材料与期望交付物,并对照来源检查输出。
npx skills add https://github.com/lilmgenius/paperthin --skill mandela

开始前需了解

  • 安装不包含模型订阅、API 额度或访问文件的权限。
  • 仓库星数和安装量反映活跃情况,不等于独立质量或安全测试。
仓库原始内容

按源仓库内容呈现,保留标题、案例、代码、表格、链接以及原文引用的演示图片。

Audit whether a validation receives independent outside ground-truth.

Goal

The Mandela Effect names a population confidently remembering something that never independently happened. A leaky validation has the same shape: designer, model, and scorer confirm each other. mandela uses eight leakage patterns to check whether a validation is independent.

Workflow

  1. Identify the validation: eval, metric, experiment, holdout, or "how we'll know". Name its components: model, scorer, designer, and dataset.
  2. Ask the core question: does external ground-truth enter independently?
  3. Test all 8 patterns below. Some apply only to particular components, such as a human subject or a scorer. Report only patterns that fire, each by name.
  4. Recall, not reason: a memorized answer recited instead of one actually derived; the system already knows the result it is supposedly computing.
  5. Wrong null hypothesis: an ablation that removes a surface label but not the underlying signal the system actually exploits, so the "control" still leaks.
  6. Shared hallucination: two components verifying each other; circularity reported as a number.
  7. Tautology: a scorer grading buckets it drew itself.
  8. Verifier = designer: a private, unreproducible recipe in a holdout's clothes.
  9. Shared-pool bias: train and holdout drawn from one labeler pool, so one bias enters both sides.
  10. Frame injection: a question that hands the subject the hypothesis.
  11. Demand characteristics: measured subjects who know they're being measured.
  12. Give the independent-ground-truth fix for each hit.

Rules

  • Blinding the output value can still leak the collection recipe.
  • Read-only: name the leak and the independence fix; don't rewrite the experiment.
  • For a high-stakes validation, you may add one independent auditor with fresh context (N=1). Give it only the validation design, without this session's reasoning, to rerun the eight patterns blind. The default remains same-session and read-only.

Verification

Turn mandela on this audit:

  1. Run patterns #3 through #5 on yourself: are you a scorer grading buckets you drew (Tautology, #4)? Is the verifier the designer (#5)? Is your verdict a shared hallucination with the design's own claims (#3)?
  2. Could a reader who didn't run the audit reach your verdict from the cited evidence alone, using independent ground-truth?
  3. The report names the root, not a laundry list.
来自同一仓库

更多 Skills

全部 Skills
lilmgenius
社区

re0-order

Move items only, reword nothing; realign a listing whose order has gone arbitrary (a list, table, catalog, sections, an enum) into a logical sequence under one stated principle. Nothing is added or removed. Use when the user names a listing, or a scope holding one, whose order no longer helps a reader follow the set, or when a pipeline the user invoked passes that scope on. Acts only inside that scope.

1小时安装量
5
GitHub Stars
1463
最近更新
11月15日
lilmgenius
社区

readback

Verify the model's understanding of a user's instruction before spending non-trivial work. Use when a request is long, bundled, high-stakes, hard to undo, or has ambiguous scope or referents such as this, that, it, the other one, whatever is cleaner, or whichever order makes sense. Restate internally, cross-check against available context, proceed silently when resolved, and surface only a genuine surviving fork.

1小时安装量
5
GitHub Stars
1463
最近更新
11月15日
lilmgenius
社区

whychk

Right after you pick an option, have a fresh outsider find the sharpest gap in the choice, then explain that gap in your own words, one gap at a time.

1小时安装量
5
GitHub Stars
1463
最近更新
11月15日
lilmgenius
社区

aim

When data lands with a thin ask (files, a repo, a paste, a chat scrollback), read it and propose what the user is trying to do with it for them to confirm. Use the moment material arrives without a fully-formed ask, or when the ask reads thinner than the data suggests.

1小时安装量
4
GitHub Stars
1463
最近更新
11月15日