| name | dr-review |
| description | Validate research data quality. Checks output files against schemas, flags NOT_FOUND clusters, incomplete entries, conflicting sources, and suspicious data. Use after a phase completes. |
You are auditing research data quality. Read the schemas and check every data file against them.
Steps
1. Load Schemas
Read .research/ARCHITECTURE.md to get the JSON schemas for each entity type. Note all required fields and valid values.
2. Audit Each Data File
For each file in data/:
Required Fields Check:
- List entries missing any required field (name, category, website, sources, status, last_updated)
- List entries where required fields are empty strings or null
Status Distribution:
- Count entries by status: COMPLETE, INCOMPLETE, INACCESSIBLE
- Flag entries with no status field
NOT_FOUND Analysis:
- Per field, count how many entries have "NOT_FOUND"
- Calculate NOT_FOUND rate per field
- Flag fields with >40% NOT_FOUND — these may need a different search strategy
Source Quality:
- Entries with zero source URLs (critical)
- Entries missing the entity's own website in sources
- Duplicate source URLs across entries (may indicate copy-paste errors)
Data Integrity:
- Entries with CONFLICTING values — list each conflict
- Entries with UNVERIFIED values — count per field
- Duplicate entries (same name or same website)
- Entries where status is COMPLETE but fields are NOT_FOUND (inconsistent)
3. Quality Report
Print findings grouped by severity:
## Data Quality Report
### Critical (must fix before report)
- 3 entries missing source URLs
- 1 entry with no status field
- 2 entries marked COMPLETE but have NOT_FOUND fields
### Warning (should investigate)
- funding_amount: 45% NOT_FOUND rate
- 4 entries with UNVERIFIED claims
- 2 potential duplicates: "Company A" appears twice
### Info
- 18 COMPLETE, 5 INCOMPLETE, 2 INACCESSIBLE
- 3 CONFLICTING values found (listed below)
- Average sources per entry: 2.3
4. Suggest Actions
Based on findings:
- Critical issues → "Fix these before running
/dr-report. Consider /dr-run to retry failed tasks."
- High NOT_FOUND rates → "Run
/dr-improve to optimize the researcher's search strategy for these fields."
- Many INCOMPLETE → "Run
/dr-run to continue. Some entities may need manual research."
- Clean data → "Data looks good. Run
/dr-report to synthesize findings."
Error Handling
- If no data files exist, tell the user to run
/dr-run first.
- If ARCHITECTURE.md is missing, tell the user to run
/dr-new first.
