Organized

docs/contact_triage_loop_draft_2026-07-09.md

@@ -0,0 +1,57 @@
+# Contact Triage Loop – Draft-Stage Contact-First Approach (2026-07-09)
+
+**Status**: Draft / exploratory. Intended as a lightweight, iterative starting point that can be tested manually in small batches before any automation is written.
+
+**Context**: This note captures the refined direction discussed on 2026-07-09 for cleaning the customer-service project (currently >1700 open tickets). The previous keyword-on-subject approach is deprioritized. The stronger signal is **contact-first review** using recent activity history.
+
+## Core Principle (Contact-First)
+- Group decisions by **contact** (author/email) rather than individual tickets.
+- For each contact, examine their **most recent 3–5 tickets/messages**.
+- Decide whether that contact (and most of their open tickets) can be safely closed as spam, stale, promotional, or low-value.
+- This directly targets the observed pattern that many spam/low-value tickets come from the same small set of repeat contacts.
+
+## Draft Workflow (Repeatable in Small Batches)
+
+1. **Discover candidate contacts**  
+   Find authors with multiple open tickets whose most recent activity is old or low-value.  
+   Starting signals:
+   - Authors with 5+ open tickets
+   - Authors whose most recent ticket is >60–90 days old
+   - Contacts whose tickets are mostly short/generic with zero follow-up
+
+2. **Per-contact review (the decision step)**  
+   For each contact, review their last 3–5 tickets and ask:
+   - Is this contact mainly sending spam, sales pitches, or unrelated noise?
+   - Has there been any meaningful reply or resolution path in the last 60–90 days?
+   - Are the tickets duplicates, test messages, or one-off low-value imports?
+   - Do they match existing `spam` or `promotional` tag patterns?
+
+3. **Close decision rules**  
+   Close most or all of a contact’s open tickets if any of the following are true:
+   - Recent pattern is clearly spam/stale/no-value
+   - No meaningful activity or resolution path in 90+ days
+   - Tickets are duplicates or low-value imports
+
+   **Do not close** if the contact shows any recent legitimate customer issue or money-related matter.
+
+4. **Execution & logging**  
+   - Use existing Redmine MCP tools or the semantic search CLI on mcpness for discovery.
+   - Apply `spam`/`promotional` tags where appropriate for future filtering.
+   - Record a short note on the contact or in a running log (which contacts were reviewed and outcome).
+
+5. **Guardrails**  
+   - Never close tickets involving money, formal complaints, or recent customer replies.
+   - When in doubt, skip the contact.
+   - After bulk cleanup, maintain a monthly review of “Open tickets from contacts with no activity in 60+ days.”
+
+## Why This Stays Draft-Stage for Now
+- Allows manual testing on the first 10–20 worst contacts to see what the real data looks like.
+- Validates the contact-history signal before investing in looping scripts or automation.
+- Directly supports shrinking the 1700+ open ticket count with low risk.
+
+## Next Natural Steps (for the other thread)
+- Run the first discovery pass on the worst offending contacts using current tools.
+- Refine the decision questions or tagging rules based on what the contacts actually show.
+- Decide whether to move any part of this into a lightweight script once the manual pattern is validated.
+
+This note is a clean hand-off for the active thread working on the contact triage loop project.