pm-meld-triage

name	pm-meld-triage
effort	low
description	Triage rules playbook for Property Meld work orders. Determines urgency, routing, and suppression rules before any action is taken.
triggers	["triage meld","classify meld","triage rules","how urgent","should I escalate"]

PM Meld Triage Playbook

Compressed operational reference. Read the message thread first — never triage from subject line alone.

Email Processing Rule (Critical)

New work orders (same-day emails, fresh meld submissions): Triage immediately — read the meld, classify urgency, post a PM message or alert David as needed. Never bulk-label and queue a fresh work order.

Old notification emails (historical backlog, activity on existing melds, same-day notifications for melds you've already processed): Apply the label configured by config.gmail_watch.processed_label_id and mark read. The matching gmail_watch.query must include -label:<processed-label-name>. No triage required.

How to distinguish: Check the email date. Same-day = immediate triage. Older = bulk label.

---

Step 0: Always Read the Thread First

Before classifying any meld, pull comments:

python3 scripts/pm-get-comments.py <meld_id>

A meld that looks unhandled from the subject may already have a vendor reply, scheduled appointment, or Brittany note in the thread. Triage on thread state, not title.

---

Urgency Classification

Level	Definition	Response
Emergency	Active safety/habitability threat	Telegram David immediately, any hour
High	No heat, sewage, lock-out, water intrusion	Escalate to an agent during day hours; wake David only if containment risk
Normal	Routine repair, appliance, cosmetic	Standard dispatch, SLA applies
Low	Cosmetic, non-functional (e.g. paint, landscaping)	Batch in next morning scan

---

Routing Rules

Nashville Melds → Brittany

Any meld tagged to a Nashville property goes to Brittany, not the standard vendor dispatch queue.

→ cortextos bus send-message an agent normal "Nashville meld <id> — routing to Brittany per protocol"

Standard Portfolio → Vendor Dispatch

All other melds follow normal vendor assignment flow.

---

Suppression Rules

Pest Control

Suppress pest control meld alerts while a vendor search is open for that meld.

• Check: does the meld have a vendor_assigned: null and a pending vendor search event?
• If yes: do not ping an agent. The search is already in flight.
• If vendor search has been open >48h with no assignment: apply RULE_R1 (see meld-ops).

Routine Follow-ups

Do not re-alert on a meld that has:

• An assigned vendor AND a scheduled date
• A Brittany note marked "handled" or "scheduled"
• A comment from Blue within the last 6h

---

Habitability Override

The following conditions bypass all suppression rules and escalate immediately:

• No heat when outdoor temp is at or below 40°F
• Active water leak with structural or unit spread risk
• No hot water >24h
• Gas smell or suspected leak
• Lock-out (tenant cannot enter unit)
• Fire or smoke (call 911 first, then David)

Override action:

cortextos bus send-telegram $CTX_TELEGRAM_CHAT_ID "URGENT: <meld_id> — <condition>. <property>. Immediate action needed."

---

Tenant SLA Targets

Priority	First Response	Resolution Target
Emergency	Immediate	4h vendor on-site
High	2h	24h
Normal	24h	5 business days
Low	72h	14 days

SLA clock starts from meld open timestamp, not from Blue's detection.

---

Age-Based Escalation Thresholds (Property Meld benchmarks)

Apply to any open meld regardless of priority, measured from meld open timestamp:

Age	Flag	Action
≥ 4 days open	Approaching critical	Include in morning scan report with age. Message an agent if no vendor assigned.
≥ 5.5 days open	Critical threshold	Flag prominently in report. Message an agent immediately regardless of time of day. Resident churn risk is active at this point.

Why 5.5 days: Property Meld data shows that repairs exceeding 5.5 days drive the probability of a positive resident review to near zero. 46% of move-outs cite maintenance as a factor.

4-day early warning leaves response time to assign, schedule, and complete before hitting the critical threshold.

---

Water Intrusion Containment Protocol

If meld description mentions: leak, flood, water coming in, ceiling drip, pipe burst, or toilet overflow:

1. Classify as High minimum (Emergency if active/spreading)
2. Pull thread immediately — check if tenant has already isolated source
3. If no isolation confirmed: message an agent to call tenant for containment steps
4. Flag for same-day vendor regardless of day/hour

---

Make-Ready Urgency

Make-ready melds (unit turnover prep) are time-sensitive when:

• Move-in date is within 5 days
• Lease is already signed

Treat as High urgency. Route to Brittany if Nashville; otherwise escalate to an agent with move-in date.

---

Decision Tree (quick reference)

Read thread
  → Already handled (vendor assigned + date set)?  → Log only, no action
  → Nashville property?                            → Route to Brittany
  → Habitability override condition?               → Telegram David immediately
  → Pest control + vendor search open?             → Suppress alert
  → Age ≥ 5.5 days?                               → Critical flag, message an agent immediately
  → Age ≥ 4 days?                                 → Approaching critical, include in report
  → Emergency priority + no vendor 4h+?            → RULE_R2: Telegram David
  → High priority?                                 → Message an agent (day hours only)
  → Normal/Low?                                    → Standard dispatch or batch

---

Copilot Threshold Logging (MANDATORY)

Before sending any recommendation to David for approval, log the decision with a full reasoning trace:

cortextos bus log-event quality blue_decision_presented info \
  --meta "{
    \"category\": \"<category>\",
    \"meld_id\": \"<meld_id>\",
    \"recommendation\": \"<one-line summary>\",
    \"safety_keywords\": [\"<keyword1>\", \"<keyword2>\"],
    \"decision_path\": [\"<rule1>:<result>\", \"<rule2>:<result>\"],
    \"confidence\": \"high|medium|low\",
    \"confidence_reason\": \"<brief reason>\"
  }"

Field Definitions

safety_keywords — list every safety-related word found in the meld description or thread. If none: [].
Examples: ["leak", "flooding"], ["no heat", "40F outside"], ["gas smell"]

decision_path — ordered list of rules checked, each as "rule:outcome".
Standard path:

"habitability_override:no|yes"
"nashville_property:no|yes"
"pest_control_suppressed:no|yes"
"age_days:<N>"
"age_flag:none|approaching_critical|critical"
"priority:emergency|high|normal|low"
"vendor_available:yes|no"

Stop the path at the first rule that terminates the decision (e.g., if habitability_override:yes, no need to log further rules).

confidence — how certain Blue is about the recommendation:

• high — clear match to existing rules, no ambiguity
• medium — rule applies but some unknown (vendor availability, scope unclear)
• low — novel situation, significant unknowns, or conflicting signals

confidence_reason — one sentence explaining confidence level. Examples:
"Trade is clear, vendor is known and available"
"Scope is unclear — resident described it vaguely, may need diagnosis visit first"
"Two rules conflict: meld age is 3 days (normal) but description mentions possible water intrusion (high)"

Category Mapping

• Dispatching in-house tech → inhouse_dispatch
• Dispatching known vendor → known_vendor_dispatch
• Dispatching new/untested vendor → new_vendor_assignment
• Lock change → lock_change
• Messaging resident → resident_comms
• Closing/canceling meld → meld_closure
• Emergency dispatch → emergency_dispatch

Log first, send recommendation second. Skipping this means the decision is invisible to an agent and your accuracy score never accumulates.

---

---

Tenant Follow-Up Escalation Ladder (3-Day Rule)

When a meld is awaiting tenant response (photos, troubleshooting answers) and no vendor is assigned:

Day	Action	Notes
Day 1	Send PM message requesting photos/response	Standard intake SOP
Day 2	Send second PM follow-up	Substitute for SMS until A2P available
Day 3	Send third PM follow-up, then escalate to David regardless of response	Surface to David as needing direct involvement
Day 4+	David's direct involvement required	Do not continue self-resolving

Tracking: Log first contact date in daily memory. Check elapsed days on each heartbeat/morning scan.

Future upgrades (automatic when available):

• When A2P SMS is live: Day 2 becomes an outbound text instead of PM message
• When calling is available: Day 3 becomes a phone call instead of PM message

Apply to: Any PENDING_ASSIGNMENT meld where Blue sent the initial photo/response request and tenant has not replied.

---

Single source of truth for triage decisions. Do not apply rules without reading the thread first.