Validation Rules

Validation Rules

Checks performed by ./scripts/validate_research.py and severity mapping for the /research-curator --validate mode.

Check Definitions

Error Severity (must fix)

• section_completeness: All 10 required sections must exist — Header, Overview, Problem Addressed, Key Statistics, Key Features, Technical Architecture, Installation & Usage, Relevance to Claude Code Development, References, Freshness Tracking
• header_fields: Header block must contain Research Date, Source URL, Version at Research, License
• empty_sections: Section heading exists but contains no content below it before the next heading

Warning Severity (should fix)

• access_dates: Every URL in the References section must have an access date in format (accessed YYYY-MM-DD) or (YYYY-MM-DD)
• freshness_tracking: Freshness Tracking section must contain Last Verified, Version at Verification, Next Review Recommended fields
• statistics_currency: Dates in Key Statistics section older than 6 months from today trigger a staleness warning
• url_format: All URLs must be valid http:// or https:// format

Info Severity (optional)

• formatting_suggestions: Minor markdown formatting issues (missing blank lines around fences, inconsistent heading levels)

Script vs Agent Responsibility

flowchart TD
    Start([Validation issue found]) --> Q{Can it be fixed mechanically?}
    Q -->|Yes — formatting, missing dates, structure| Script[Script reports issue with file:line]
    Q -->|No — needs fresh research, content generation| Agent[Spawn @research-curator with --fix flag]
    Script --> Report[Include in JSON output for orchestrator]
    Agent --> Report

Script handles detection only — it identifies issues and reports them with file path, line number, severity, and message.

Agent handles fixes that require:

• Gathering fresh statistics (re-research)
• Writing missing section content
• Updating stale references with current URLs
• Refreshing version numbers from upstream

Orchestrator decides which issues to auto-fix vs report to user based on severity.

JSON Output Schema

{
  "summary": {
    "total": "number of entries scanned",
    "passed": "entries with zero errors",
    "errors": "total error-severity issues",
    "warnings": "total warning-severity issues"
  },
  "entries": [
    {
      "file": "category/filename.md (relative to research/)",
      "status": "pass | fail",
      "issues": [
        {
          "check": "check name from definitions above",
          "severity": "error | warning | info",
          "message": "human-readable description",
          "line": "line number or null"
        }
      ]
    }
  ]
}