agentleFS
TrecekTrecek/useful-claude-skills/.claude/skills/exp-lens-sensitivity-robustness/SKILL.md

exp-lens-sensitivity-robustness

Skill1 starsChanged 8 months agoMITView on GitHub

Create Sensitivity & Robustness experimental design analysis identifying load-bearing analytic choices and untested perturbations. Robustness lens answering "Which assumptions are load-bearing?

The file

We split this skill into the rules it gives an agent

12 rules. About 0 are common, 10 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈10

    Distinctive · estimate

    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:

    Do not litter the codebase with useless comments, TODO markers, or explanatory annotations — the skill output and diagram speak for …

    line 30
  2. Distinctive:

    Treat "untested" as equivalent to "robust"

    line 31
  3. Very specific:

    BEFORE creating any diagram, LOAD the /mermaid skill using the Skill tool - this is MANDATORY

    line 38
  4. Distinctive:

    Look for: choice, option, default, parameter, threshold, method, alternative

    line 50
  5. Distinctive:

    Look for: ablation, without, remove, disable, vary, sweep, drop

    line 54
  6. Distinctive:

    Look for: normalize, tokenize, augment, crop, resize, filter, clean, impute

    line 58
  7. Distinctive:

    Look for: learning_rate, batch_size, epochs, dropout, hidden_size, temperature, alpha, beta

    line 62
  8. Distinctive:

    Look for: shift, domain, transfer, cross, out_of_distribution, generalize, different

    line 66
  9. Distinctive:

    Untested: No evidence either way

    line 85
  10. Distinctive:

    If a diagram adds value, create a simplified flowchart.

    line 89
  11. Very specific:

    Write the analysis to: temp/exp-lens-sensitivity-robustness/exp_diag_sensitivity_robustness_{YYYY-MM-DD_HHMMSS}.md

    line 104
  12. Distinctive:

    Before creating the diagram, verify:

    line 211

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “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

Sensitivity & Robustness Experimental Design Lens

Philosophical Mode: Robustness Primary Question: "Which assumptions are load-bearing?" Focus: Ablation Structure, Preprocessing Sensitivity, Metric Sensitivity, Hyperparameter Sensitivity, Distribution Shift

When to Use

  • Results may depend on specific preprocessing choices
  • Need to verify robustness of conclusions across conditions
  • Ablation study seems incomplete or cherry-picked
  • User invokes /exp-lens-sensitivity-robustness or /make-experiment-diag sensitivity

Critical Constraints

NEVER:

  • Modify any source code or experiment files
  • Do not litter the codebase with useless comments, TODO markers, or explanatory annotations — the skill output and diagram speak for themselves
  • Treat "untested" as equivalent to "robust"

ALWAYS:

  • Build a full sensitivity matrix (choices x perturbation types)
  • Classify every analytic choice as load-bearing, minor, or untested
  • Flag cases where the most impactful choices are the least tested
  • Distinguish between ablations that were run and choices that were simply fixed
  • BEFORE creating any diagram, LOAD the /mermaid skill using the Skill tool - this is MANDATORY

Analysis Workflow

Step 1: Launch Parallel Exploration Subagents

Spawn Explore subagents to investigate:

Analytic Choices Made

  • Find all decision points in the analysis pipeline
  • Look for: choice, option, default, parameter, threshold, method, alternative

Ablation Coverage

  • Find which factors have been ablated
  • Look for: ablation, without, remove, disable, vary, sweep, drop

Preprocessing Variations

  • Find preprocessing steps that could be done differently
  • Look for: normalize, tokenize, augment, crop, resize, filter, clean, impute

Hyperparameter Sensitivity

  • Find which hyperparameters were tuned vs fixed
  • Look for: learning_rate, batch_size, epochs, dropout, hidden_size, temperature, alpha, beta

Distribution/Environment Variations

  • Find evidence of testing under different conditions
  • Look for: shift, domain, transfer, cross, out_of_distribution, generalize, different

Step 2: Build the Sensitivity Matrix

Rows = analytic choices. Columns = perturbation types (remove, change, stress).

For each cell: Does the conclusion survive the perturbation?

Step 3: Classify Analytic Choices

CRITICAL — Analyze Assumption Load: For every analytic choice:

  • What happens if this choice were made differently?
  • Is there evidence from ablations, sweeps, or prior literature?
  • Are the most impactful choices the least tested?

Classify each choice as:

  • Load-bearing: Conclusion changes if this choice changes
  • Minor: Conclusion is robust to changes in this choice
  • Untested: No evidence either way

Step 4: Create the Optional Perturbation Diagram

If a diagram adds value, create a simplified flowchart. This is OPTIONAL for this hybrid lens — the tables are the primary output.

Direction: TB (choices flow down through perturbation to conclusion stability)

Subgraphs: "ANALYTIC CHOICES", "PERTURBATIONS TESTED", "CONCLUSION STABILITY"

Node Styling:

  • stateNode class: analytic choice nodes
  • handler class: perturbation type nodes
  • output class: stable conclusion nodes
  • gap class: load-bearing untested choice nodes
  • detector class: sensitivity threshold nodes

Step 5: Write Output

Write the analysis to: temp/exp-lens-sensitivity-robustness/exp_diag_sensitivity_robustness_{YYYY-MM-DD_HHMMSS}.md


Output Template

# Sensitivity & Robustness Analysis: {Experiment Name}

**Lens:** Sensitivity & Robustness (Robustness)
**Question:** Which assumptions are load-bearing?
**Date:** {YYYY-MM-DD}
**Scope:** {What was analyzed}

## Sensitivity Matrix

| Analytic Choice | Remove | Change Value | Stress Test | Overall Classification |
|----------------|--------|-------------|-------------|------------------------|
| {choice name} | Stable / Fragile / Untested | Stable / Fragile / Untested | Stable / Fragile / Untested | Load-bearing / Minor / Untested |

## Load-Bearing Assumptions

| Assumption | Evidence Type | Impact if Changed | Tested? |
|------------|--------------|-------------------|---------|
| {assumption} | Ablation / Sweep / Literature / None | High / Medium / Low | Yes / No |

## Ablation Coverage Assessment

| Factor | Ablated? | Result | Interpretation |
|--------|----------|--------|----------------|
| {factor name} | Yes / No | {delta metric if yes} | Conclusion holds / Fragile / Unknown |

## Robustness Profile

| Dimension | Status | Notes |
|-----------|--------|-------|
| Preprocessing choices | Robust / Fragile / Untested | {detail} |
| Hyperparameter choices | Robust / Fragile / Untested | {detail} |
| Metric choices | Robust / Fragile / Untested | {detail} |
| Distribution shift | Robust / Fragile / Untested | {detail} |

## Perturbation Diagram (Optional)

```mermaid
%%{init: {'flowchart': {'nodeSpacing': 50, 'rankSpacing': 60, 'curve': 'basis'}}}%%
flowchart TB
    %% CLASS DEFINITIONS %%
    classDef cli fill:#1a237e,stroke:#7986cb,stroke-width:2px,color:#fff;
    classDef stateNode fill:#004d40,stroke:#4db6ac,stroke-width:2px,color:#fff;
    classDef handler fill:#e65100,stroke:#ffb74d,stroke-width:2px,color:#fff;
    classDef phase fill:#6a1b9a,stroke:#ba68c8,stroke-width:2px,color:#fff;
    classDef newComponent fill:#2e7d32,stroke:#81c784,stroke-width:2px,color:#fff;
    classDef output fill:#00695c,stroke:#4db6ac,stroke-width:2px,color:#fff;
    classDef detector fill:#b71c1c,stroke:#ef5350,stroke-width:2px,color:#fff;
    classDef gap fill:#ff6f00,stroke:#ffa726,stroke-width:2px,color:#000;
    classDef integration fill:#c62828,stroke:#ef9a9a,stroke-width:2px,color:#fff;

    subgraph Choices ["ANALYTIC CHOICES"]
        CHOICE1["Preprocessing Choice<br/>━━━━━━━━━━<br/>{e.g., normalization}"]
        CHOICE2["Model Choice<br/>━━━━━━━━━━<br/>{e.g., architecture}"]
        LOAD_BEARING["Load-Bearing Choice<br/>━━━━━━━━━━<br/>{untested critical choice}"]
    end

    subgraph Perturbations ["PERTURBATIONS TESTED"]
        PERTURB1["Remove<br/>━━━━━━━━━━<br/>Ablation"]
        PERTURB2["Change Value<br/>━━━━━━━━━━<br/>Sensitivity sweep"]
        THRESHOLD["Sensitivity Threshold<br/>━━━━━━━━━━<br/>{delta that changes conclusion}"]
    end

    subgraph Stability ["CONCLUSION STABILITY"]
        STABLE["Stable Conclusion<br/>━━━━━━━━━━<br/>{holds under perturbation}"]
    end

    CHOICE1 --> PERTURB1
    CHOICE2 --> PERTURB2
    PERTURB1 --> THRESHOLD
    PERTURB2 --> THRESHOLD
    THRESHOLD --> STABLE
    LOAD_BEARING -.->|"untested risk"| THRESHOLD

    class CHOICE1,CHOICE2 stateNode;
    class LOAD_BEARING gap;
    class PERTURB1,PERTURB2 handler;
    class THRESHOLD detector;
    class STABLE output;

Color Legend:

ColorCategoryDescription
TealAnalytic ChoicesDecision points in the pipeline
YellowLoad-Bearing UntestedCritical choices with no perturbation evidence
OrangePerturbationsTypes of tests applied
RedSensitivity ThresholdsPoints where conclusion may change
Dark TealStable ConclusionsResults robust to perturbation

Recommendations

  1. {Most urgent ablation to run — highest impact untested choice}
  2. {Preprocessing sensitivity test needed}
  3. {Distribution shift or domain generalization test needed}

---

## Pre-Diagram Checklist

Before creating the diagram, verify:

- [ ] LOADED `/mermaid` skill using the Skill tool
- [ ] Using ONLY classDef styles from the mermaid skill (no invented colors)
- [ ] Diagram will include a color legend table

---

## Related Skills

- `/make-experiment-diag` - Parent skill for lens selection
- `/mermaid` - MUST BE LOADED before creating diagram
- `/exp-lens-estimand-clarity` - For clarifying which conclusions are being stress-tested
- `/exp-lens-iterative-learning` - For tracking robustness improvements across experiment iterations

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.