docs: INSTRUMENTS.md and PERFORMANCE.md state the bench gate at 5% over five workloads; it is 0.5% over ten

未关闭 适合新手
  • 哪里不对？
#2,282 1 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
2/5
预计耗时
1-3 小时
新手友好度
72/100
Issue 类型
文档
描述清晰度
描述清楚
活跃度
活跃
技术栈
markdown, shell
领域
documentation, performance

调研方向

从 bench/check_regression.sh（第 16 行，THRESHOLD_PCT）和 ls bench/*.eigs 入手，确认实际阈值和 10 个程序，然后与 codspeed.yml 对比。更正 docs/INSTRUMENTS.md 第 130 行以及 docs/PERFORMANCE.md 第 19、93 和 228-229 行中的数字。当文档给出 --vs 的 0.5% 阈值、53 倍的比率、10 个程序中有 4 个不在该通道中，以及 29 个自测用例时，即完成。

由索引模型根据 Issue 内容生成。

描述

area:docs found-by:duet-audit kind:docs-drift

Found by audit area 10 (performance claims), read-only, against main b367cd7.

The register is supposed to be the record of each device's resolution, and two of its numbers now describe an older gate. Each item below was checked against the tree.

Where Says Tree says
docs/INSTRUMENTS.md:130 (CodSpeed row, "Decides") "the required bench job (cachegrind Ir, THRESHOLD_PCT=5, a different instrument) 526 times" the worst A/A spread bench/check_regression.sh:16 THRESHOLD_PCT=0.5 for the --vs mode CI runs (#2072, 2026-10-28). 0.5% / 0.0095% is about 53 times, not 526. (Row 127, the bench row itself, is correct.)
docs/INSTRUMENTS.md:130 ("Population and blind spots") "13 rows (3 of 9 bench/*.eigs programs are not in the lane)" ls bench/*.eigs gives 10; codspeed.yml has 6 of them, so 4 are not in the lane (alloc_heavy, both conservative_*, observer_gate_soft_keyword).
docs/PERFORMANCE.md:19 the bench job "fails on more than a few percent" 0.5% (--vs); 5% only for the local file mode.
docs/PERFORMANCE.md:228-229 the bench job checks "five workloads"; codspeed.yml "declares the same five workloads" The job measures every bench/*.eigs (10); the lane declares 6 bench/ rows (it added unobserved_armed_loop).
docs/PERFORMANCE.md:93 test_string_scaling.sh --selftest # 23 cases == selftest 29 run, 29 passed == (matches the INSTRUMENTS.md row's plants: 29).

Why it matters: the CodSpeed row's ratio is the register's argument that both gates sit far above their resolution. With the real threshold it is 53 times, still comfortably above, but the register should carry the measured number, not one from before #2072.

Done when

  • docs/INSTRUMENTS.md:130 gives the --vs threshold as 0.5% with its ratio to the worst A/A spread, and the lane's population as 4 of 10 bench/*.eigs programs absent.
  • docs/PERFORMANCE.md states the 0.5% --vs threshold, the real workload counts for the job and the lane, and 29 self-test cases.
  • Optional, to stop the next drift: tools/docs_claims_check.sh or performance_observer_docs.py derives these counts and the threshold from the tree instead of prose (the front-door rule already bans derived counts in README/llms.txt).

— Duet (orchestrator)

主要语言
C
星标
3
派生
7
平均合并
6 小时 28 分钟
30 天内合并 PR
312

环境准备

在 Codespaces 中打开

在浏览器里用你自己的 GitHub 账号启动这个项目的开发容器。

从这里开始

  1. 先读完整个 Issue，再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库，在一个分支上完成修改。
  4. 提交 Pull Request，并在描述里引用这个 Issue 编号。

InauguralSystems/EigenScript 的其他 Issue

查看 InauguralSystems/EigenScript 的全部 Issue

相似的 Issue

更多 C Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。