agentleFS
Sign inStart free
wilbeibiwilbeibi/wilbeibi-skills/skills/benchmark-checklist/SKILL.md

benchmark-checklist

Skill2 starsChanged 2 months agoMITView on GitHub

Check whether a measured performance claim supports a comparison or decision. Use when running benchmarks, reporting a measured speedup or regression, or choosing between implementations from timing results. Do NOT use for interpreting published benchmarks without running them or for general debugging.

42 other projects use a version of this.

Adopted by

Most recently taken up by

Used with

Projects with it also use ketchup-video, principle-explain-the-number and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • tristanphvnSkill

    ketchup-video

    Catch the user up like /keel:ketchup, with a short silent video of slides for what happened and the waiting calls as text under it. Use for /keel:ketchup-video.

    Also in 14 projects that use this

  • ianclearySkill

    principle-explain-the-number

    Apply before you trust, report, or act on a number you measured: a speedup, a regression, a throughput, a latency, or an eval result. Find what limits it, and rule out that it measured something other than the work you think.

    Also in 28 projects that use this

  • ianclearySkill

    poteto-help

    Guides users through pstack setup, /poteto-mode, and picking the skill, playbook, or principle for a task. Type /poteto-help with a question.

    Also in 14 projects that use this

  • instructureSkill

    ketchup

    Rebuild context after /clear by analyzing uncommitted changes across Canvas LMS and all plugin repositories

    Also in 25 projects that use this

  • drafaelSkill

    in-your-own-words

    Restate the user's goals and the problem they are trying to solve to check understanding. Manual invocation only; use when the user explicitly invokes in-your-own-words or requests this understanding check. Do not solve the task or rewrite the assistant's last answer.

    Also in 24 projects that use this

  • jaydubya818Skill

    principle-attack-the-premise

    Apply when repeated fixes fail and a shared assumption may be wrong.

    Also in 27 projects that use this

The file

We split this skill into the rules it gives an agent

2 rules. 0 are common, 2 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 40 edited versions we read keep these.

  • 2

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Test the sentence the measurement is supposed to support.

    line 8
  2. Distinctive:

    Write the proposed claim with its metric, workload, and scope.

    line 12

“Common” is measured: kept by at least 6 in 10 of the 40 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

benchmark-checklist

Test the sentence the measurement is supposed to support.

Before measuring

Write the proposed claim with its metric, workload, and scope. Read the harness to see what it times, counts, awaits, and ignores. Define a correctness check independently of the timing result.

Inspect machine load and resource limits. If competing work cannot be avoided, interleave the alternatives and report the interference; do not stop unrelated processes.

Check the result

QuestionRequired evidence
What limits it?Profile separately from reported timing runs. Map the bottleneck to code or a resource; check that the load generator is not the bottleneck.
Is the comparison fair?Comparable versions, data, builds, cache state, concurrency, and production settings. Tune avoidable configuration bottlenecks before recommending an implementation. If tuning is out of scope, bound the claim to the configurations tested and leave the adoption decision open.
Is the magnitude possible?Check bandwidth, core capacity, and the affected share of total time. Removing 10% of a run saves at most 10% of its time, about a 1.11x speedup. A result beyond a physical or arithmetic bound needs investigation.
Did it succeed?Count failures, retries, timeouts, and non-success responses. Validate actual outputs; fast rejection is not fast completion.
Does it repeat?For a comparison, start with at least five runs per side in alternating order, unless the user sets a smaller budget. Record warmup, raw results, median, and range. Mark the conclusion inconclusive when the budget or variation cannot support the claimed difference.
Does it affect the user?Measure the relevant end-to-end path with realistic data and concurrency. State a microbenchmark's share of the whole; without that check, keep the claim at microbenchmark scope.
Did the work happen inside the timer?Observe completion counts or effects: consumed iterators, awaited work, returned bytes, written rows, or used results. Setup, asynchronous tail work, caching, and no-ops can invalidate the timed region.

For an explicitly requested ballpark, one run is enough. Still verify successful output and work inside the timer, label it a single run, and do not turn it into a winner recommendation. A choice between alternatives is a comparison.

Decision

  • Faster / slower: the successful, comparable runs support a difference beyond observed variation; explain the limiting work.
  • No measurable difference: the valid experiment cannot distinguish the alternatives at its observed resolution.
  • Inconclusive: correctness, completion, comparability, repetitions, or the proposed explanation is unresolved. Name the missing evidence.

Distinguish elapsed-time reduction from throughput increase. For example, 100 ms to 80 ms is 20% less time and, for serial work, 1.25x throughput. Preserve units and the statistic being compared.

Output

Lead with the verdict and its scope. Give the main before/after number with units, run count, median and range, correctness/work counts, and bottleneck evidence. Link the harness and raw results when available. State limits that could change the decision; do not generalize a configuration result into an implementation ranking.

Adapted from pstack/benchmark-checklist, under the MIT license.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free