Plan ablation studies, experiments that remove or replace parts of a machine learning method to test their contribution. Define reviewer questions, expected outcomes, priorities, execution order, and compute estimates, then implement experiments and record findings.
Use when main results pass result-to-claim (claim_supported=yes or partial) and ablation studies are needed for paper submission.
Review the instructions the agent will receive before you install.
Systematically design ablation studies that answer the questions reviewers will ask. Codex leads the design (reviewer perspective), CC reviews feasibility and implements.
/result-to-claim with claim_supported = yes or partial/auto-review-loop reviewer identifies missing ablationsCC reads available project files to build the full picture:
idea-stage/docs/research_contract.md, legacy docs/research_contract.md, or project CLAUDE.md)mcp__codex__codex:
model: gpt-6-astra
config: {"model_reasoning_effort": "xhigh"}
prompt: |
You are a rigorous ML reviewer planning ablation studies.
Given this method and results, design ablations that:
1. Isolate the contribution of each novel component
2. Answer questions reviewers will definitely ask
3. Test sensitivity to key hyperparameters
4. Compare against natural alternative design choices
Method: [description from project files]
Components: [list of removable/replaceable components]
Current results: [key metrics from experiments]
Claims: [what we claim and current evidence]
For each ablation, specify:
- name: what to change (e.g., "remove module X", "replace Y with Z")
- what_it_tests: the specific question this answers
- expected_if_component_matters: what we predict if the component is important
- priority: 1 (must-run) to 5 (nice-to-have)
Also provide:
- coverage_assessment: what reviewer questions these ablations answer
- unnecessary_ablations: experiments that seem useful but won't add insight
- suggested_order: run order optimized for maximum early information
- estimated_compute: total GPU-hours estimate
Normalize Codex response into structured format:
## Ablation Plan
### Component Ablations (highest priority)
| # | Name | What It Tests | Expected If Matters | Priority |
|---|------|---------------|---------------------|----------|
| 1 | remove module X | contribution of X | performance drops on metric Y | 1 |
| 2 | replace X with simpler Z | value of learned vs fixed | drops, especially on dataset A | 2 |
### Hyperparameter Sensitivity
| # | Parameter | Values to Test | What It Tests | Priority |
|---|-----------|---------------|---------------|----------|
| 3 | lambda | [0.01, 0.1, 1.0] | sensitivity to regularization | 3 |
### Design Choice Comparisons
| # | Name | What It Tests | Priority |
|---|------|---------------|----------|
| 4 | joint vs separate matching | whether joint adds value | 4 |
### Coverage Assessment
[What reviewer questions these ablations answer]
### Unnecessary Ablations
[Experiments that seem useful but won't add insight — skip these]
### Run Order
[Optimized for maximum early information]
### Estimated Compute
[Total GPU-hours]
Before running anything, CC checks:
ablation-no-module-X)what_it_tests and expected_if_component_matters. No "just try it" experiments.No reviews yet. Tell others what you thought.
No login needed. Share how it worked for you, ask a question, or note where it did not run.
No comments yet. Be the first to write one.
Plan and run ablation studies, experiments that remove or replace parts of a method to test their contribution. Organize reviewer questions, priorities, expected outcomes, and compute costs, then record results and findings.
Understand a single arXiv paper through an AlphaXiv summary, with full text and LaTeX source available when needed. Use it to identify key findings or ask specific questions about methods, experiments, and equations.
Compare machine learning experiment results from JSON or CSV files. Calculate improvements over a baseline, summarize variation across repeated runs, identify trends and unusual results, and suggest experiments to test possible explanations.
Search arXiv papers by topic or paper ID, review authors and abstracts, and summarize contributions stated in the abstracts. Download selected papers or all search results as PDFs to build a local paper library.
Improve a compiled LaTeX paper through two rounds of Claude review, edits, and recompilation. The workflow addresses inconsistent reasoning, unsupported claims, unclear writing, and visual presentation, while keeping a cumulative log of changes.
Improve an existing LaTeX paper through two rounds of Gemini review, targeted edits, and PDF rebuilding. Check whether claims match the evidence, resolve notation and presentation issues, and retain review logs and PDF versions.