quality-gate

Quality Gate: Paper ↔ Analysis Consistency

Cross-check every numerical claim in the paper against analysis output files. Reports only — never edits.

Input: $ARGUMENTS — path to the paper draft, or leave blank to auto-detect.

Read ${CLAUDE_PLUGIN_ROOT}/rules/quality-gates.md before assigning the final verdict and severity.

Step 1: Locate the Paper Draft

If $ARGUMENTS is provided, use that path. Otherwise glob for:

• manuscripts/**/*.tex
• manuscripts/**/*.qmd
• manuscripts/**/*.md

If multiple drafts found, use AskUserQuestion to let the user pick:

• header: "Paper"
• question: "Multiple drafts found. Which manuscript should I check?"
• options: list up to 4 found files (label: filename, description: path and file size). If more than 4, group by directory.

Step 2: Extract Numerical Claims

Read the full manuscript and extract every quantitative claim:

• Coefficients and standard errors — e.g., "The effect is 0.23 (SE = 0.04)"
• Sample sizes — e.g., "N = 4,521 observations"
• Percentages and proportions — e.g., "42% of firms..."
• Means, medians, ranges — e.g., "average income of $45,000"
• Table and figure references — e.g., "Table 2 shows...", "as seen in Figure 1"
• p-values and significance statements — e.g., "statistically significant at the 1% level"

Record location (section, paragraph, line number if available) for each claim.

Step 3: Inventory Output Files

Glob for all output files:

• output/tables/**/*.tex — regression and summary tables
• output/tables/**/*.html — HTML versions
• output/figures/**/*.pdf, output/figures/**/*.png — figures
• output/**/*.rds, output/**/*.pkl, output/**/*.parquet, output/**/*.csv — saved objects

Build an inventory with file paths and sizes.

Step 4: Dispatch Verifier Agent

Dispatch the verifier agent via Task to perform the heavy verification work. Pass it:

• The full list of numerical claims extracted in Step 2 (with locations)
• The output file inventory from Step 3
• The paper draft path
• The bibliography file path