list-dedup

List Dedup

A pure data-processing utility: dedup and clean BEFORE any ColdIQ enrichment. This is the single biggest credit saver — every duplicate you remove is a paid find/enrich call you don't make. See resources/credit-optimization.md. This skill makes no API calls.

When to dedup

• Before enrichment (never pay to enrich the same person twice)
• After combining multiple sources (database search + enrichment + ad scrape)
• After scoring (collapse duplicates, keep the highest-scored row)

Primary key: LinkedIn URL (normalize first)

import re
def normalize_linkedin(url):
    if not url: return ""
    return re.sub(r'\?.*$', '', url).rstrip('/').lower().replace('https://www.', 'https://')

def dedup_people(rows):
    seen, out = set(), []
    for r in rows:
        key = normalize_linkedin(r.get('linkedin_url', '')) or (r.get('email','').lower())
        if key and key not in seen:
            seen.add(key); out.append(r)
    return out

Secondary key: company name / company LinkedIn URL

Normalize company names (strip Inc/LLC/Ltd suffixes, lowercase) or, better, dedup on the company LinkedIn URL when present.

Multi-source combine pattern

1. Normalize columns — map each source's columns to one standard schema (full_name, title, linkedin_url, email, company_name, company_domain, company_linkedin, source, score).
2. Load all source CSVs.
3. Dedup people by normalized LinkedIn URL (fallback email); companies by company LinkedIn / name.
4. Sort & export — Tier 1 first, then by persona, then company name.

Per-company contact caps

Cap 5–10 contacts per company so you don't blast one account. Rank by title priority (C-level/VP/Director > Manager > IC), keep the top N per company_domain.

Cross-reference against TAM

Left-join the contact list against the scored TAM on company_domain to attach Tier, and drop contacts at DQ'd companies before enrichment.

Checklist

• Normalize LinkedIn URLs before comparing
• Dedup people by LinkedIn URL (fallback email)
• Dedup companies by LinkedIn URL / normalized name
• Map all sources to one schema before combining
• Keep highest-scored row on duplicate
• Apply per-company contact cap (5–10)
• Cross-reference against TAM, drop DQ companies
• THEN enrich (see contact-enrichment)