research-grounded

To audit architecture decisions for missing research provenance, scan IMPL.md and ROADMAP.md for committed library, framework, and pattern selections lacking `# research(YYYY-MM):` evidence tags. The audit classifies decisions and reports non-destructive gaps grouped by file, ready for selective confirmation.

A documentation audit for design decisions identifies committed architecture bets stated as fact without provenance. You need this audit when planning documents use decision language for library or pattern selections without `# research(YYYY-MM):` evidence tags, preventing costly architecture mistakes from ungrounded claims.

Yes, you can verify architecture bets with web search through an optional best-practice grounding loop. It detects un-grounded decisions, performs web searches, and proposes adding correct research provenance entries, ensuring only verified sources are added without inventing citations.

Yes, the provenance gap audit works with multi-file planning directories by applying fan-out analysis. It uses grep-seeded candidate detection to find decision language across files, ensuring all committed architecture and design decisions are checked for missing research evidence.

The best way to add research provenance is through decision-focused reporting that outputs one actionable gap per genuine decision. After detecting un-grounded decisions, a web-grounding loop proposes adding a correct `# research(YYYY-MM):` provenance entry with verified sources for selective confirmation.