agentleFS
jskhermanjskherman/engg-skills/skills/design-of-experiments/SKILL.md

design-of-experiments

Skill7 starsChanged 6 months agoApache-2.0View on GitHub

Generate full-factorial and two-level factorial DOE plans with randomization support and compute main-effect estimates. Use when planning a small experimental campaign with 2-5 factors and a clear single response. Don't use for response-surface methodology (RSM is outside the scope), Plackett-Burman screening (different algorithm), or process-data observational study design (use process-causal-inference-dags).

4 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

6 rules. 0 are common, 4 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 4

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Add replication externally if you need pure-error estimates or curvature checks.

    line 67
  2. Distinctive:

    After the run, compute main effects.

    line 68
  3. Distinctive:

    If main effects are statistically significant, iterate to RSM with a different tool.

    line 69
  4. Very specific:

    If the number of factors exceeds 5, switch to fractional-factorial or Plackett-Burman screening (use pyDOE2 outside this skill).

    line 90
  5. Very specific:

    Run the listed script with representative inputs and an --output file when a deterministic calculation is available.

    line 97
  6. Distinctive:

    Check result magnitudes against the stated assumptions, references, and a hand calculation or known operating range before reporting them.

    line 99

“Common” is measured: kept by at least 6 in 10 of the 4 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

Design of Experiments

Overview

Factorial DOE helpers:

  • Full factorial design enumeration for an arbitrary number of factors at arbitrary level counts.
  • Two-level factorial design (2^k) with optional randomization.
  • Main-effect estimates from a fitted run (simple averaging contrast).

Suitable for hands-on screening campaigns with a small number of factors; not a replacement for full RSM (central composite, Box-Behnken) or optimal-design software.

Prerequisites

  1. uv available.

When to Use

  • Planning a 2-5 factor screening campaign with clear inputs and a single response.
  • Generating a randomised run sheet for a small experiment.
  • Computing the rough magnitude of main effects after a small run.

Don't use for

  • Response-surface methodology (RSM, central composite, Box-Behnken).
  • Plackett-Burman or fractional-factorial confounding analysis.
  • Observational data — DOE assumes you control the factor levels.
  • Mixture experiments (constrained simplex designs).

Utility Scripts

  • uv run scripts/doe.py full-factorial --factor "T=300,320" --factor "P=1e5,2e5" --output /tmp/fact.json
  • uv run scripts/doe.py two-level --factors "A,B,C" --randomize --seed 42 --output /tmp/2k.json

Procedure

  1. List factors and their levels (2 for screening, more for full factorial).
  2. Choose the design (full factorial for small k; two-level if you only want main effects and 2-factor interactions).
  3. Randomise the run order to mitigate time-order effects.
  4. Add replication externally if you need pure-error estimates or curvature checks.
  5. After the run, compute main effects.
  6. If main effects are statistically significant, iterate to RSM with a different tool.

Pitfalls

  • Choosing two levels that are too close; main effects fall in the noise.
  • Forgetting to randomise; time-order confounds the factor effects.
  • Reporting effects without replicate variability.
  • Confusing main effect (averaged over the other factors) with conditional slope (computed at fixed values of the other factors).
  • Treating a two-level design as a curvature test; this script does not add centre points or run an RSM analysis.
  • Using a 2^k design when you have only one shot per condition (no replicates) and treating ±SE as meaningful.
  • Picking response that is correlated with multiple physical phenomena without thinking about what the design is testing.
  • Forgetting blocking when the experiment spans multiple days / raw-material lots.

Fallback Strategies

  • If the number of factors exceeds 5, switch to fractional-factorial or Plackett-Burman screening (use pyDOE2 outside this skill).
  • For curvature exploration, switch to a central composite design with a dedicated package.

Verification

  • Run the listed script with representative inputs and an --output file when a deterministic calculation is available.
  • Confirm the JSON result contains ok: true, expected units, and no unhandled warnings.
  • Check result magnitudes against the stated assumptions, references, and a hand calculation or known operating range before reporting them.

References

  • references/factorial_designs.md — design conventions and analysis outline.
  • Montgomery, Design and Analysis of Experiments (any edition).

Anti-Patterns

  • Reporting "the effect of T is X" without naming the level range and the other factors held fixed.
  • Designing an experiment without randomization and reporting the estimates as if independent.
  • Using 2-level factorial for a known curved response.

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.