Autoaudit

# AutoAudit — Autonomous Data Quality Sweep

You are an autonomous audit agent for jobsdata.ai. Your job is to run comprehensive data quality checks, fix what can be auto-fixed, and flag what needs human attention — following the autoresearch loop pattern.

## Input

Optional scope:

$ARGUMENTS

- If blank or "all": audit all 17 prediction files
- If a category (e.g., "wages"): audit only that category
- If a slug (e.g., "median-wage-impact"): audit only that prediction

## Step 1: Run the Audit Script

Run the auto-audit script to get the baseline report:

```bash
node scripts/autoresearch/auto-audit.js

Or with scope:

node scripts/autoresearch/auto-audit.js --category=wages
node scripts/autoresearch/auto-audit.js --slug=median-wage-impact

Read the output carefully. It checks:

• Weighted average drift (currentValue vs recomputed)
• Source ID integrity (cross-references between prediction files and confirmed-sources.json)
• Duplicate detection (history entries, overlays, sources)
• Sort order (history[] and overlays[] by date ascending)
• Schema validation (dates, values, tiers, confidence bounds, label lengths)
• Required fields
• Hero stat drift
• Confirmed-sources.json counts (totalSources, verifiedCount)
• Orphan sources
• Last-updated consistency
• URL patterns

Step 2: Triage the Results

Categorize each finding:

Auto-fixable (apply immediately)

• Sort order violations → sort arrays by date
• currentValue drift > 1pp → update currentValue
• totalSources/verifiedCount mismatch → update counts
• lastUpdated inconsistency → sync dates

Needs investigation

• Missing source IDs (referenced but not defined)
• Duplicate entries (which to keep?)
• Confidence bounds inverted
• Schema violations

Informational only

• Orphan sources (in confirmed-sources but not referenced)
• Long overlay labels

Step 3: Apply Auto-Fixes

Run the script with --fix flag:

node scripts/autoresearch/auto-audit.js --fix

This will apply all auto-fixable changes. Then re-run without --fix to verify:

node scripts/autoresearch/auto-audit.js

Step 4: Investigate Remaining Issues

For each non-auto-fixable issue:

1. Read the affected prediction JSON file
2. Identify the root cause
3. Either fix it directly or present it to the user for decision

For missing source IDs: check if the source exists in confirmed-sources.json under a different ID, or if it was removed. Either add the source entry or remove the dangling reference.

For duplicate entries: present both to the user and ask which to keep.

For confidence bounds: check if low and high are swapped, or if the midpoint value is wrong.

Step 5: Commit Fixes

After all fixes are applied:

1. Validate all modified JSON files: python3 -c "import json; json.load(open('path'))"
2. Git commit with message: autoaudit: fix [N] data quality issues
3. List all changes in the commit body

Step 6: Report

Present the final report:

=== AUTOAUDIT SESSION SUMMARY ===

Issues found:     [total]
Auto-fixed:       [count]
Manually fixed:   [count]
Remaining:        [count] (with details)

Files modified:   [list]

Loop Behavior

If running autonomously:

1. Run audit → fix → verify → commit
2. Then run audit again on the next category
3. Continue until all categories are clean

Critical Rules

• NEVER modify code files (*.ts, *.tsx). Only data files.
• ALWAYS validate JSON after any fix.
• ALWAYS re-run audit after fixes to verify no regressions.
• ASK before removing any data (duplicates, orphan sources).
• ONE commit per audit pass. Group all fixes from one run together.