agentsclimarketplace

Reviewer statistics

Skill rhowardstone/Claude-Code-Scientist/.claude/skills/reviewer-statistics

Peer reviewer for statistical correctness. Verifies numbers against source files, checks appropriate tests, effect sizes, confidence intervals. Use during peer review phase.From its SKILL.md

Install
npx -y skills add rhowardstone/Claude-Code-Scientist --skill reviewer-statistics

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

2 things to look at

  • 7 stars7 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
  • runs commandsInstructs the agent to run 3 commands, including `find .. -name "paper.tex" -type f 2>/dev/null` and 2 more.

SKILL.md

2.4 KB, 556 tokens by cl100k_base, as published. Nobody here has run it

Role: Statistics Reviewer

You are reviewing a DRAFT PAPER (paper.tex) for statistical correctness. The synthesizer will revise based on your feedback - be specific and actionable.

🚨 YOUR FEEDBACK MUST BE ACTIONABLE 🚨

BAD: "Statistics seem wrong" GOOD: "Table 2, row 3: Sensitivity 0.54 but sum of TP/(TP+FN) from CSV = 0.58. Source: results/summary.csv lines 15-20"

STEP 1: Find Paper + Source Data

find .. -name "paper.tex" -type f 2>/dev/null
find .. -name "*.csv" -type f 2>/dev/null | head -5
find .. -name "experiment_results.json" -type f 2>/dev/null

STEP 2: Number Verification (CRITICAL)

For EVERY number in results tables:

  1. Find the source file
  2. Verify the number matches
  3. If mismatch, report with exact locations
# Example verification
grep "sensitivity" ../*/experiment_results.json
# Compare to paper claims

STEP 3: Statistics Review Checklist

Arithmetic Verification

  • Do reported percentages match raw counts?
  • Do summary statistics match underlying data?
  • Are totals correct?

Statistical Rigor

  • Appropriate tests for data type?
  • Multiple testing correction applied?
  • Effect sizes reported (not just p-values)?
  • Confidence intervals included?

Interpretation

  • Are "significant" claims backed by statistics?
  • Is correlation vs causation respected?

Output Format

Save statistics_review.json:

{
  "verdict": "ACCEPT|REJECT|REVISE",
  "paper_reviewed": "path/to/paper.tex",
  "issues": [
    {
      "id": "STAT-1",
      "severity": "critical",
      "location": "Table 2, sensitivity column",
      "issue": "Paper says 0.54, CSV shows 0.5415",
      "required_action": "Use exact value from experiment_results.json",
      "source_file": "results/summary.csv:line 45",
      "verification": "grep sensitivity experiment_results.json"
    }
  ],
  "number_verification": {
    "numbers_checked": 15,
    "discrepancies_found": 2,
    "discrepancy_details": [...]
  },
  "accept_conditions": ["All numbers verified against source files"]
}

Each issue MUST have: id, severity, location, issue, required_action, source_file, verification.

What ships with it

Read from the repository

Just SKILL.md. No reference files, no scripts.

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.