GroundForce Capital Management (GFCM)—

Floating "💬 Send feedback to Ricki" button on both dashboard views → modal (optional company + message) → POST /api/ai-summit-list/feedback. Persists to new Postgres table gf_dashboard_feedback AND posts to Slack #call-notes via the ARK bot token (channel overridable via env GF_FEEDBACK_SLACK_CHANNEL). Removes Kathryn as middleman — Dan flags issues straight to Ricki. Verified end-to-end: DB insert + live Slack delivery confirmed.

• Feedback channel moved from #call-notes → #rickis-agent (C0B41CCESBB) per Ricki. GF_FEEDBACK_SLACK_CHANNEL default updated in app.py; tested live.
• List too loose (holding cos, real estate, hotels, B2B in the kept buckets). Name-matching can't fix it ("Suntory Holdings"=beverage vs "G1 Partners"=fund). Built data/build_summit_relevance.py — Haiku judges each kept-bucket company as a real consumer brand or not (rubric: keep CPG/food&bev/health&wellness/beauty/DTC + consumer restaurant chains; drop holdings/PE/VC, real estate/homebuilders, hotels/resorts/travel, B2B mfg/packaging/distributors, agencies/IT, medical/pharma, nonprofits). Output data/summit-relevance.json.
• Wired as a 2nd filter layer in _is_summit_target (mtime-cached, auto-applies live; manual industry override still wins). ~28% drop rate → projected ~2,650 (from 3,694).
• OPEN DECISION: restaurants/food-service chains currently KEPT (per the call). If Ricki wants pure product brands only, change rubric + rerun.