Skip to content
Back to skills

Pyperformance Result Compare

ASecurity

Use when 比较已有 pyperformance run.json 或 speedup.json,判断收益、回归与噪声;不启动跑分。

  • 2 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added September 20, 2026
documentationpythonperformance

Security analysis

A100/100

Scanned September 20, 2026

npx -y skills add sisibeloved/cpython-optimize-skill --skill pyperformance-result-compare --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Pyperformance Result Compare?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Pyperformance Result Compare
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/sisibeloved-pyperformance-result-compare/badge)](https://www.skillsdirectory.com/skills/sisibeloved-pyperformance-result-compare)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: pyperformance-result-compare
description: Use when 比较已有 pyperformance run.json 或 speedup.json,判断收益、回归与噪声;不启动跑分。
---

# pyperformance Result Compare

负责解释结果,不负责跑 benchmark。

## 输入

- baseline `run.json`
- candidate `run.json`
- `speedup.json`
- 提交 baseline / candidate commit
- 口径 baseline:CPython JIT、CinderX JIT、解释执行等

## 判断

- 命令口径是否一致
- 按 `../using-cpython-optimize/references/baseline-source-contract.md` 核对 baseline source 是否 `baseline_source_verified`,包括口径 baseline、提交 baseline、source path、commit/ref、dirty 状态、`patchlevel.h`、`SOABI` 和容器 bind mount。
- 按 `../using-cpython-optimize/references/pyperformance-affinity-guidance.md` 核对 baseline/candidate 的原始/实际 `--affinity`、可用 CPU 映射和并行/串行口径是否一致。
- 先按 `../using-cpython-optimize/references/pyperformance-env-contract.md` 核对 baseline/candidate 的 `--inherit-environ`、driver/worker env 和唯一差异轴。
- CinderX JIT 口径必须核对 worker 内证据:`.pth`、`pyvenv.cfg` / `include-system-site-packages`、`import cinderx` / `_cinderx`、`cinderx.__file__`、`cinderx.get_import_error()`、`cinderx.is_initialized()`。
- baseline/candidate 是否只在目标变量上不同
- 方差、噪声和异常值
- 收益范围、无收益范围、未验证范围
- 需要回到 `pyperformance-worker-run` 的异常用例
- 如果 baseline source、affinity 口径、环境契约、worker JIT 证据缺失或 baseline/candidate 不一致,先降级结论,不把 `run.json` 写成可信性能收益。

## 反问 Gate

- 无法确定 baseline/candidate 配对或比较含义时,询问缺失事实;不凭文件名猜测实验轴。
- baseline source、worker 或 affinity 证据缺失时照实报告受限观察与补证要求,不把它当可信收益。
- 不稳定数据直接标为噪声或证据不足。已有数据分析不启动远程补跑;只有用户要求补测或扩大收益声明时,按已有授权与预算决定运行范围。

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…