⏱️
AI EngineeringPython

Auditable Method Choice

by tangxiangru

Auditable Method Choice is an AI Engineering skill for Claude Code, published by tangxiangru in AutoR.

804 stars25 forkson tangxiangru/AutoRAdded 2026/08/22Repository updated 2026/08/22
agentaiai-scientistauto-researchclaudeclaude-codecliharnessllmopenaipaperscience
Install in seconds
Install Auditable Method Choice
Copy Auditable Method Choice into your Claude Code skills folder. Run the command in your terminal, or review the source on GitHub before installing.
terminal
npx degit https://github.com/tangxiangru/AutoR/tree/main/src/skills/a-model-you-can-audit-is-not-a-model-that-scores ~/.claude/skills/a-model-you-can-audit-is-not-a-model-that-scores

Requires Node.js. Downloads this skill only — not the rest of the repository — into your Claude Code skills folder.

Without Node.js

git clone https://github.com/tangxiangru/AutoR.git

Clones the whole repository, then copy the skill’s own directory into your skills folder yourself.

In this catalog

Source file
src/skills/a-model-you-can-audit-is-not-a-model-that-scores/SKILL.md in tangxiangru/AutoR
Installs to
~/.claude/skills/a-model-you-can-audit-is-not-a-model-that-scores
Collection
One of 25 skills cataloged from this repository
Category
AI Engineering2451 skills

What Auditable Method Choice does

Auditable Method Choice helps you decide whether a more complex method fits the remaining clock by timing a small run, extrapolating full cost, and writing a go/no-go before you commit. It also covers checkpointing and split/seed control so the stronger option can fail safely.

Auditable Method Choice is cataloged under AI Engineering on DirSkills. Auditable Method Choice comes from a repository tagged agent, ai, ai-scientist, auto-research and claude.

Documentation

README

The auditable method wins the argument and loses the score

Given a fixed clock and a scored predictions file, there is a recurring choice between two methods:

  • one you can build in twenty minutes, cross-validate cleanly, explain fully, and defend against every question a reviewer asks;
  • one the field actually uses to get the published number, which needs an hour of setup you have not done, has failure modes you cannot fully enumerate, and might not finish.

Every incentive inside a rigour-checking pipeline points at the first. The gates reward an auditable choice. The reviewer is easier to satisfy. The write-up is cleaner. And on a benchmark that grades predictions, none of that is measured.

This is the opening of the README. Read the full README on GitHub.

Frequently asked about Auditable Method Choice

  • What else does tangxiangru publish alongside Auditable Method Choice?

    Auditable Method Choice is one of 25 skills that DirSkills catalogs from tangxiangru/AutoR, the repository it ships in. Its siblings there include A Deliverable Is Not an Instruction, A Value You Did Not Measure Still Has A Source and Answer The Why, Not Only The What. Each one is a separate skill with its own page in this directory, installs the same way Auditable Method Choice does, and is maintained by tangxiangru in that same repository. The rest of the collection is listed on the tangxiangru/AutoR page.

  • How does Auditable Method Choice compare to other AI Engineering skills?

    Auditable Method Choice ranks #1591 by stars among the 2451 AI Engineering skills in this catalog. The most-starred ones next to it are Architecture Decision Records, AI-First Engineering and Agentic OS. DirSkills ranks by the star count of the repository each skill ships in, so that order reflects how popular those repositories are rather than any review of Auditable Method Choice against them. Open each page to compare what they document and how they install.

More from tangxiangru/AutoR

Auditable Method Choice is one of 25 skills cataloged on DirSkills from tangxiangru/AutoR.

See all 25 skills
🧭
1w ago

A Deliverable Is Not an Instruction

A Deliverable Is Not an Instruction helps separate research deliverables from harness instructions when building a study plan and checking coverage. Use it to keep `report_plan.json` focused on findings and to mark genuinely unreachable items honestly.
AI Engineering
80425
📝
1w ago

A Value You Did Not Measure Still Has A Source

A Value You Did Not Measure Still Has A Source explains how to handle deliverables this run cannot measure or produce, using attributed published values instead of fabricating or omitting them. It covers where to place cited values and how to compare them to your own results.
Writing
80425
📝
1w ago

Answer The Why, Not Only The What

Answer The Why, Not Only The What helps you write results and discussion sections that explain why an effect happens, not just whether it happened. Use it when a reviewer or task asks for mechanism claims, competing explanations, or limits on what the data support.
Writing
80425
🧪
1w ago

Assume This Stage Is The Last One You Get

Assume This Stage Is The Last One You Get advises planning each early stage of a timed research run so it leaves a valid model, score, and rerunnable script if the later stages never happen. It is used when wall-clock limits make deferred work unlikely to execute.
AI Engineering
80425
🔭
1w ago

Astronomy Caption Specification

Astronomy Caption Specification turns figure captions into a plotting specification for reproducing astronomy figures when the rendered source is unavailable. Use it to keep panel order, series, colors, references, normalization, and error bars aligned with the paper.
Writing
80425
🔭
1w ago

Astronomy Error Budget Audit Trail

Astronomy Error Budget Audit Trail helps you document uncertainty propagation, fit bookkeeping, and residual diagnostics before quoting a result. Use it when a measurement depends on calibration chains, covariance, or a model fit.
Writing
80425