blindspot

name	blindspot
description	Peripheral vision audit — find what the author has become blind to in their results. Run in a FRESH TERMINAL.

Blindspot Audit

Based on Scott Cunningham's Blindspot skill (MixtapeTools). The Shklovsky Principle: by the time you've spent months on a paper, the main finding has collapsed your attention. Blindspot makes the stone stony again.

CRITICAL: Fresh Terminal Required

This skill MUST be run in a new Claude Code session with no prior context from the authoring sessions. The separation is what makes it work — the auditing Claude has no prior commitments to the author's narrative.

Steps

1. Read the paper in paper-N/paper/main.tex or the latest draft
2. Read all output: tables in output/tables/, figures in output/figures/
3. Read the strategy memo: pre-analysis-plan/strategy-memo.md
4. Read the codebook: data/codebook/variables.md, data/codebook/sample.md
5. Apply the Blindspot Grid (four quadrants):

The Blindspot Grid

	Things Present in Output	Things Absent from Output
Vices (problems)	Anomalies normalized — outliers not questioned, coefficient signs rationalized, patterns dismissed as noise	Checks never run — subgroups not split, placebo tests missing, robustness not attempted
Virtues (opportunities)	Richness ignored — heterogeneity patterns, mechanism evidence in appendix tables, interesting coefficients on controls	Extensions not considered — natural next questions the data can answer, adjacent literatures not connected

6. For each finding, document:

   • What was found (specific table/figure/coefficient)
   • Which quadrant it falls in
   • Why the author likely missed it
   • Recommended action (investigate, discuss, add, remove)
   • Severity (critical / important / minor)

7. File the report in paper-N/correspondence/blindspot/YYYY-MM-DD_blindspot-report.md

Output Format

# Blindspot Report — Paper [N]
Date: YYYY-MM-DD

## Vices — Problems Hiding in Plain Sight
### [Finding 1]
- **Location:** Table 3, Column 4
- **Issue:** [Description]
- **Why missed:** [Author likely focused on...]
- **Action:** [Investigate / discuss / remove]
- **Severity:** Critical

## Vices — Checks Never Run
### [Finding 1]
...

## Virtues — Richness Being Ignored
### [Finding 1]
...

## Virtues — Extensions Not Considered
### [Finding 1]
...

## Summary
[N] findings: [X] critical, [Y] important, [Z] minor.

Important

• NEVER modify author files — read only
• File the report in correspondence/blindspot/ — this is an immutable record
• The author responds in a separate response file