Fewshot Attempt 03

HarborCRM Skill

Environment

Base URL: GDPEVO_ENV_BASE_URL (supplied by runner). All API calls use this as the root. Never use localhost unless the env var points there.

API Endpoints Reference

Always start by fetching /api/policies — it contains follow-up offset days, campaign codes, lead opportunity amounts, priority-tier amount mappings, platform inclusion lists, and other configuration.

Task Type: Event Reconciliation

Used when the task asks for post-event CRM handoff involving sponsors, badge scans, and lead qualification.

Sponsor Status Classification

For each sponsor order/package, check invoices to determine status:

• paid_deferred: An invoice exists for the sponsor AND paid_amount >= package_amount (fully paid).
• open_invoice: An invoice exists AND 0 < paid_amount < package_amount (partially paid). Report open_balance = package_amount - paid_amount.
• proposal_only: No invoice exists for this sponsor package. invoice_id is null, paid_amount is 0.

Only include active sponsor records. Skip canceled/inactive sponsor packages.

Revenue Totals

Sum package_amount by status into sponsor_revenue_totals (integer USD). For open_invoice, include a separate open_invoice_balance sum of all open_balance values.

Lead Qualification from Badge Scans

A badge scan is a qualified non-sponsor lead when ALL of:

1. The contact's company is NOT a sponsor company
2. The badge type is a business badge (exclude: student, press, exhibitor-only, staff, etc.)
3. The CRM account is NOT already disqualified
4. The badge is for an active attendee (not canceled)

Exclusion Reasons (badge-level)

CRM Action Determination

For each qualified lead, cross-reference CRM accounts and contacts:

• Account: If an account with matching name/ID exists → update_existing. Otherwise → create_account.
• Contact: If a contact with matching normalized email exists in CRM → update_existing. Otherwise → create_contact.
• Campaign member: If not already a campaign member for this event → add_campaign_member.

Lead Opportunity Amount

Use the event's lead opportunity amount from the event data or policies. All qualified non-sponsor leads get this same amount. lead_pipeline_total = sum of all qualified leads' opportunity amounts.

Follow-up Dates

Computed from the event's end date plus policy-defined offset days:

• Lead follow-up due date: event_end_date + lead_followup_offset_days (from policy)
• Sponsor finance due date: event_end_date + sponsor_finance_offset_days (from policy)

Sponsor Finance Follow-up

sponsor_finance_accounts lists account names of sponsors with status open_invoice or proposal_only (unpaid/partially-paid), sorted ascending. sponsor_finance_task_count = length of this list.

CRM Action Counts

Count each CRM action across all qualified leads:

• accounts_create, accounts_update
• contacts_create, contacts_update
• campaign_members_create, campaign_members_update

Badge-Only Contact Normalization

For contacts that appeared only via badge scan (no CRM contact record), normalize:

• email: lowercase, trim. Empty string "" if not provided.
• phone: digits only (strip all non-digit characters). Empty string "" if not provided.

Task Type: Trade Show Prospecting

Used when the task asks for qualified exhibitor lists from a trade show for a campaign.

Qualification

An exhibitor qualifies when their business type is an OEM/manufacturer of platforms covered by the campaign policy. Exclude:

• distributor_only — company is a distributor, not a manufacturer
• service_only — company provides services only
• sensor_vendor_only / sensor_only — company only sells sensors/components
• research_only — academic/research institution

Platform Matching

Platforms must be from the controlled enum: AUV, ROV, Underwater Camera. Sort platform lists in this enum order. Only include platforms the exhibitor actually covers that are also in the campaign's target platform set.

Priority Tier Assignment

Based on demo request and meeting interest score:

• A: requested_demo == true AND interest_score >= 90
• B: requested_demo == true AND interest_score >= 80
• C: all other qualified leads

Opportunity Sizing

Map priority tier to USD amount using policy or task-specified mappings:

• A → 120000
• B → 90000
• C → 50000

CRM Action (Trade Show)

• If the exhibitor's company has an existing CRM account → update_existing, include crm_account_id
• If no CRM account exists → create_account, crm_account_id is null
• Excluded exhibitors → no_import

Ranking (when specified)

Sort qualified leads by:

1. requested_demo == true first (demo requests before non-demo)
2. interest_score descending
3. Platform count descending (more platforms = broader coverage = higher rank)
4. company_name ascending (alphabetical tiebreaker)

Assign contiguous 1-based rank.

Aggregate Counts

• qualified_total: count of qualified exhibitors
• excluded_count / excluded_near_misses_total: count of excluded exhibitors
• platform_counts: per-platform count across qualified exhibitors
• priority_counts: per-tier count (A, B, C) — include 0 for empty tiers
• existing_crm_overlap_count: count of qualified leads with existing CRM accounts
• existing_crm_overlap_account_ids: their account IDs, sorted ascending
• total_estimated_opportunity_usd: sum of all opportunity estimates

Task Type: Batch Import

Used when preparing raw import contacts for CRM ingestion.

Deduplication

Group raw contacts by normalized email (lowercase, trimmed). Within each group, pick one winner:

• Prefer the row with the most complete contact data (has both email AND phone)
• If tied, prefer the earliest captured_at timestamp
• If still tied, prefer lowest row_id lexicographically

Removed duplicates are tracked in duplicate_summary with key format email:{normalized_email}.

Suppression

Check each raw contact against the suppression list. A contact is suppressed if their normalized email matches a suppression entry. Suppressed contacts go to removal_summary with reason suppressed.

Unusable Contacts

A contact is unusable (missing_contact) if it has no email AND no phone after normalization.

Removal Summary

• unusable_removed_count: count of missing_contact rows
• suppressed_removed_count: count of suppressed rows
• removed_rows: all removed rows (duplicates, suppressed, unusable) sorted by row_id ascending

CRM Action for Clean Contacts

For each surviving clean contact:

• If company matches an existing CRM account → update_existing, set existing_account_id
• If no CRM account match → create_account, existing_account_id is null
• existing_contact_id: the matching CRM contact ID if email matches, otherwise null

Import Action Totals

Count across ALL raw contacts (not just clean):

• create_account: clean contacts needing new accounts
• update_existing: clean contacts matching existing accounts
• no_import: rows removed as unusable (missing_contact)
• suppress: rows removed as suppressed

Campaign Member Count

campaign_member_import_count = number of clean contacts (surviving after dedup, suppression, and unusable removal).

General Rules

Data Normalization

• Email: Always lowercase, trim whitespace. Use "" (empty string) when no email provided — never null for normalized fields.
• Phone: Strip ALL non-digit characters. US/international leading digits are preserved. Use "" when no phone provided.
• Dollar amounts: Always integer USD. No decimal places, no cents.

Sorting Conventions

• Sponsor lists: by account_name ascending (case-sensitive, lexicographic)
• Qualified lead account lists: by account_name ascending
• Excluded records: by company_name ascending, then contact_name ascending
• Badge decisions: by badge_id ascending
• Campaign member actions: by subject_key ascending
• Clean contacts: by clean_contact_id ascending
• Duplicate keys: by key ascending
• Removed rows: by row_id ascending
• Platform lists within items: in enum order (AUV, ROV, Underwater Camera)
• Account ID lists: ascending

Null vs Empty

• Use null for object references that don't exist (e.g., invoice_id, existing_account_id, existing_contact_id, crm_account_id)
• Use "" (empty string) for missing normalized text fields (normalized_email, normalized_phone, email, phone)
• Use null for exclusion_reason when the item is NOT excluded

Controlled Vocabularies

Always use the exact string values from the answer template or listed in this skill. Do not invent new statuses, reasons, or action names.

Workflow Order

1. Fetch policies first — they define offsets, thresholds, and configuration
2. Fetch all relevant entity endpoints in parallel (event/show, CRM, finance)
3. Cross-reference: match badges to accounts, invoices to sponsors, contacts to CRM
4. Classify each record (sponsor status, lead qualification, exclusion reason)
5. Compute aggregates and follow-up data
6. Sort all lists per the sorting rules
7. Output JSON conforming to the answer template — no extra keys, no prose outside JSON